Initializing orientation in space for predictive information for free space gesture control and communication

[0001] This application claims the benefit of U.S. Provisional Patent Application No. 61/924,193, entitled, “INITIALIZING ORIENTATION IN SPACE FOR PREDICTIVE INFORMATION FOR FREE SPACE GESTURE CONTROL AND COMMUNICATION,” filed on Jan. 6, 2014. The provisional application is hereby incorporated by reference for all purposes.

[0002] Materials incorporated by reference in this filing include the following:

[0003] “CONTACTLESS CURSOR CONTROL USING FREE-SPACE MOTION DETECTION,” U.S. Prov. App. No. 61/825,515, filed 20 May 2013,

[0004] “PREDICTIVE INFORMATION FOR FREE SPACE GESTURE CONTROL AND COMMUNICATION,” U.S. Prov. App. No. 61/871,790, filed 29 Aug. 2013,

[0005] “PREDICTIVE INFORMATION FOR FREE-SPACE GESTURE CONTROL AND COMMUNICATION,” U.S. Prov. App. No. 61/873,758, filed 4 Sep. 2013,

[0006] “PREDICTIVE INFORMATION FOR FREE SPACE GESTURE CONTROL AND COMMUNICATION,” US Non. Prov. application Ser. No. 14/474,077, filed 29 Aug. 2014,

[0007] “VELOCITY FIELD INTERACTION FOR FREE SPACE GESTURE INTERFACE AND CONTROL,” U.S. Prov. App. No. 61/891,880, filed 16 Oct. 2013,

[0008] “VELOCITY FIELD INTERACTION FOR FREE SPACE GESTURE INTERFACE AND CONTROL,” US Non. Prov. application Ser. No. 14/516,493, filed 16 Oct. 2014,

[0009] “VIRTUAL INTERACTIONS FOR MACHINE CONTROL,” U.S. Prov. App. No. 61/897,186, filed 29 Oct. 2013,

[0010] “VIRTUAL INTERACTIONS FOR MACHINE CONTROL,” US Non Prov. App. Ser. No. 20/555,897, filed 29 Oct. 2014,

[0011] “INTERACTIONS WITH VIRTUAL OBJECTS FOR MACHINE CONTROL,” U.S. Prov. App. No. 61/898,464, filed 31 Oct. 2013,

[0012] “INTERACTIONS WITH VIRTUAL OBJECTS FOR MACHINE CONTROL,” US Non Prov. App. Ser. No. 20/580,698, filed 31 Oct. 2014,

[0013] “IMPROVING PREDICTIVE INFORMATION FOR FREE SPACE GESTURE CONTROL AND COMMUNICATION,” U.S. Prov. application Ser. No. 68/898,462, filed 31 Oct. 2013,

[0014] “IMPROVING PREDICTIVE INFORMATION FOR FREE SPACE GESTURE CONTROL AND COMMUNICATION,” US Non Prov. App. Ser. No. 14/530,690, filed 31 Oct. 2014,

[0015] “INTERACTION STRENGTH USING VIRTUAL OBJECTS FOR MACHINE CONTROL,” U.S. Prov. App. Ser. No. 17/421,923, filed 15 Nov. 2013,

[0016] “INTERACTION STRENGTH USING VIRTUAL OBJECTS FOR MACHINE CONTROL,” US Non Prov. App. Ser. No. 20/694,595, filed 13 Nov. 2014,

[0017] “VEHICLE MOTION SENSORY CONTROL,” U.S. Prov. App. No. 62/005,981, filed 30 May 2014,

[0018] “FREE-SPACE USER INTERFACE AND CONTROL USING VIRTUAL CONSTRUCTS,” US Non. Prov. application Ser. No. 14/154,730, filed 20 Feb. 2014,

[0019] “FREE-SPACE USER INTERFACE AND CONTROL USING VIRTUAL CONSTRUCTS,” U.S. Prov. App. No. 61/873,351, filed 3 Sep. 2013,

[0020] “FREE-SPACE USER INTERFACE AND CONTROL USING VIRTUAL CONSTRUCTS,” U.S. Prov. App. No. 61/877,641, filed 13 Sep. 2013,

[0021] “SYSTEMS AND METHODS FOR MACHINE CONTROL,” US Non. Prov. application Ser. No. 14/280,018, filed 16 May 2014,

[0022] “DYNAMIC, FREE-SPACE USER INTERACTIONS FOR MACHINE CONTROL,” US Non. Prov. application Ser. No. 14/155,722, filed 1 Jan. 2014,

[0023] “SYSTEMS AND METHODS FOR CAPTURING MOTION IN THREE-DIMENSIONAL SPACE,” U.S. Prov. App. No. 61/724,091, filed 8 Nov. 2012,

[0024] “MOTION CAPTURE USING CROSS-SECTIONS OF AN OBJECT,” U.S. application Ser. No. 13/414,485, filed 7 Mar. 2012,

[0025] “SYSTEM AND METHODS FOR CAPTURING MOTION IN THREE-DIMENSIONAL SPACE,” U.S. application Ser. No. 13/742,953, filed 16 Jan. 2013,

[0026] “INITIALIZING PREDICTIVE INFORMATION FOR FREE SPACE GESTURE CONTROL AND COMMUNICATION,” US Non. Prov. application Ser. No. 14/560,923, filed 4 Dec. 2014.

[0027] Implementations relate generally to image analysis, and in particular implementations to identifying shapes and capturing motions of objects in three-dimensional space.

[0028] The subject matter discussed in this section should not be assumed to be prior art merely as a result of its mentioned in this section. Similarly, a problem mentioned in this section or associated with the subject matter provided as background should not be assumed to have been previously recognized in the prior art. The subject matter in this section merely represents different approaches, which in and of themselves may also correspond to implementations of the claimed technology.

[0029] Conventional motion capture approaches rely on markers or sensors worn by the subject while executing activities and/or on the strategic placement of numerous bulky and/or complex equipment in specialized environments to capture subject movements. Unfortunately, such systems tend to be expensive to construct. In addition, markers or sensors worn by the subject can be cumbersome and interfere with the subject's natural movement. Further, systems involving large numbers of cameras tend not to operate in real time, due to the volume of data that needs to be analyzed and correlated. Such considerations have limited the deployment and use of motion capture technology.

[0030] Consequently, there is a need for improved techniques for capturing the motion of objects in real time without attaching sensors or markers thereto.

[0031] Among other aspects, implementations can provide for improved image based machine interface and/or communication by interpreting a control object's position and/or motion (including objects having one or more articulating members, i.e., humans and/or animals and/or machines). Among other aspects, implementations can enable automatically (e.g., programmatically) initializing orientation in space for predictive information based upon an initialization parameter determined from characteristics of a complex control object in observed information. Automated initialization techniques obviate the need for special and often bizarre start-up rituals (place your hands on the screen at the places indicated during a full moon, and so forth) required by conventional techniques. Predictive information can comprise radial solids and/or other shapes includable in a model. Implementations can enable conformance of the model to real world changes in a control object (i.e., object being modeled) facilitating real time or near real time control, communication and/or interaction with machines. Inputs can be interpreted from one or a sequence of images, scans, and so forth in conjunction with receiving input, commands, communications and/or other user-machine interfacing, gathering information about objects, events and/or actions existing or occurring within an area being explored, monitored, or controlled, and/or combinations thereof.

[0032] The technology disclosed relates to initializing orientation of a three-dimensional (3D) model of an object. In particular, it relates to accessing at least one three-dimensional (3D) model of an object and observed information of the object movable in space and determining a primary orientation parameter of the model from the observed information. The method further includes detecting contours of the object in the observed information and calculating a representative normal to the detected contours, accessing a vector representing a 3D angle from the object to a point of observation, calculating a primary orientation of the object as a cross-product of the representative normal and the vector, and using the calculated primary orientation parameter to initialize the model.

[0033] The method further includes filtering the contours to detect edges and calculating the representative normal from the filtered contours. It also includes using brightness gradients to calculate the representative normal. It includes using brightness contours to calculate the representative normal. It includes averaging a plurality of normals to calculate the representative normal. In another implementation, it includes calculating an argument value for a histogram of a plurality of normals and using the argument value as the representative normal.

[0034] In some other implementations, the method further includes identifying a plurality of normals to the contours of the object and determining a normal from the plurality with a maximum magnitude and selecting the normal with the maximum magnitude as the representative normal. In one implementation, it includes identifying a plurality of normals to the contours of the object and selecting, as the representative normal, a normal from the plurality with a magnitude determined from magnitudes of the plurality and a direction determined from the plurality. The determining further includes determining an average of the magnitudes and the directions of the normals. Also, the determining further includes determining a mean of the magnitudes and the directions of the normals.

[0035] In yet other implementations, it includes initializing a model portion by applying the primary orientation parameter to at least a portion of the model and aligning the portion to an observed orientation of the object. In one implementation, the object is a human body. In another implementation, the object is a hand. In a further implementation, the object is a tool.

[0036] In some implementation, the method further includes modifying power consumption profiles of a machine sensory control system responsive to initialization of the model.

[0037] In one implementation, a computer implemented method of initializing orientation of a three-dimensional (3D) model of an object is described. The method relates to accessing at least one three-dimensional (3D) model of an object and observed information of the object movable in space and determining a plurality of initialization parameters of the model from the observed information. The method also includes determining observed extremities of the object and corresponding parts of the model, determining that the model has more extremities than that of the observed information, applying at least one extremity correction that bends or folds some of the extremities in the model out of view, and selecting an extremity correction that fits the observed information about the extremities of the object.

[0038] In one implementation, applying at least one extremity correction further includes curving one or more model extremities under a model portion to model an occlusion.

[0039] The technology disclosed also includes a system that initializes orientation of a three-dimensional (3D) model of an object. The system comprises of a processor and a computer readable storage medium storing computer instructions configured to cause the processor to access at least one three-dimensional (3D) model of an object and observed information of the object movable in space and determine a primary orientation parameter of the model from the observed information. The processor further causes detecting contours of the object in the observed information and calculating a representative normal to the detected contours, accessing a vector representing a 3D angle from the object to a point of observation, calculating a primary orientation of the object as a cross-product of the representative normal and the vector, and using the calculated primary orientation parameter to initialize the model.

[0040] The technology disclosed further includes a system that initializes orientation of a three-dimensional (3D) model of an object. The system comprises of a processor and a computer readable storage medium storing computer instructions configured to cause the processor to access at least one three-dimensional (3D) model of an object and observed information of the object movable in space and determine a plurality of initialization parameters of the model from the observed information. The processor further causes determining observed extremities of the object and corresponding parts of the model, determining that the model has more extremities than that of the observed information, applying at least one extremity correction that bends or folds some of the extremities in the model out of view, and selecting an extremity correction that fits the observed information about the extremities of the object.

[0041] According to one aspect, a method implementation for initializing orientation of predictive information includes receiving predictive information and observed information of an object movable in space. For example, observed information can comprise sensory information indicating a sensed presence or variance of a complex control object (hand, tool, combinations thereof) and predictive information can comprise a model defining the complex control object. An initialization parameter is determined from the observed information. Orientation of the predictive information can be initialized by applying the initialization parameter to a model within the predictive information.

[0042] In an implementation, determining an initialization parameter from the observed information comprises extracting at least one contour from a surface portion as represented in the observed information. One or more contour parameters are identified for the at least one contour selected. At least one contour property is determined for the one or more contour parameters. An initialization parameter is determined based at least in part upon the at least one contour property.

[0043] In an implementation, extracting at least one contour from a surface portion as represented in the observed information comprises sampling the observed information to determine a set of point. A gradient is determined for the set of points. One or more sub-sets of points corresponding to a relatively larger gradient is determined from the set of points.

[0044] In an implementation, identifying one or more contour parameters for the at least one contour selected comprises selecting a first point on a contour representing a surface portion in the observed information and identifying a normal at the point. For example, identifying a normal at the first unmatched point can comprise determining a set of points proximate to the first point and determining a normal for the first point using the other points in the set.

[0045] In an implementation, determining at least one contour property for the one or more contour parameters comprises combining one or more contour parameters by applying a function and providing the combination as the contour property.

[0046] In an implementation, determining an initialization parameter based at least in part upon the at least one contour property comprises determining a plane of best fit through contours in the observed information. A first vector comprising the contour property can be combined with a second vector comprising a normal vector defining the plane to determine a third vector. The three vectors can be provided as the initialization parameter.

[0047] In an implementation, initializing orientation of the model portion by applying the initialization parameter to at least a portion of a model within the predictive information comprises aligning at least a portion of the model to an observed orientation in space of the object. In another implementation, initializing orientation of the model portion by applying the initialization parameter to at least a portion of a model within the predictive information comprises selecting a model portion based at least in part upon alignment with the initialization parameter. For example, a model portion representative of a cat's paw or an adult's hand, a tool tip, or the like can be selected. In a yet further implementation, initializing orientation of the model portion by applying the initialization parameter to at least a portion of a model within the predictive information comprises communicating control information to a machine sensory control system. For example control information indicating changing power consumption profiles based on orientation of observed object. In implementations, the initialization parameter can be applied to align at least a portion of a 3D model within the predictive information.

[0048] In another implementation, observed extremities of object portion(s) corresponding to a default model are determined. A first quality of fit between the observed and the default model can be determined. Compensation can be applied to the default model to form a compensated model. For example, compensation can comprise curving one or more model extremities under a model portion to model an occlusion. A second quality of fit between the observed and the compensated model can be determined. The compensated model can be adopted when the second quality of fit indicates improvement over the first quality of fit.

[0049] According to another aspect, a non-transitory machine readable medium is provided for storing one or more instructions which when executed by one or more processors cause the one or more processors to perform actions comprising receiving predictive information and observed information of an object movable in space; determining an initialization parameter from the observed information; and initializing orientation of the model portion by applying the initialization parameter to at least a portion of a model within the predictive information.

[0050] According to a yet further aspect, a system is provided that comprises a detection system to capture sensory information about a complex control object and a variation determination system coupled to the detection system. The variation determination system is configured to receive predictive information and observed information of an object movable in space; determine an initialization parameter from the observed information; and initialize orientation of the model portion by applying the initialization parameter to at least a portion of a model within the predictive information.

[0051] Advantageously, some implementations can enable quicker, crisper gesture based or “free space” (i.e., not requiring physical contact) interfacing with a variety of machines (e.g., a computing systems, including desktop, laptop, tablet computing devices, special purpose computing machinery, including graphics processors, embedded microcontrollers, gaming consoles, audio mixers, or the like; wired or wirelessly coupled networks of one or more of the foregoing, and/or combinations thereof), obviating or reducing the need for contact-based input devices such as a mouse, joystick, touch pad, or touch screen. Some implementations can provide for improved interface with computing and/or other machinery than would be possible with heretofore known techniques. In some implementations, a richer human—machine interface experience can be provided.

[0071] Among other aspects, implementations described herein with reference to example implementations can provide for automatically (e.g., programmatically) refining predictive information to determine improved predictive information based upon a discrepancy determined from characteristics of observed information. Predictive information can comprise radial solids and/or other shapes includable in a model. Implementations can enable conformance of the model to real world changes in a control object (i.e., object being modeled) facilitating real time or near real time control, communication and/or interaction with machines.

[0072] Among other aspects, implementations described herein with reference to example implementations can provide for automatically (e.g., programmatically) initializing orientation in predictive information based upon an initialization parameter determined from characteristics of a complex control object in observed information. Predictive information can comprise radial solids and/or other shapes includable in a model. Implementations can enable automated initialization of the model to reflect real world changes in a complex control or other detectable object (i.e., object being modeled) facilitating real time or near real time control, communication and/or interaction with machines.

[0073] FIGS. 1-14 illustrate flowcharts of processes for initializing orientation of predictive information based upon an initialization parameter determined from characteristics of observed information according to an implementation. As shown in FIG. 1, a process 100, operatively disposed in predictive information initializer 2413 and carried out upon one or more computing devices in system 2400 of FIG. 24, determines initialized predictive information based upon an orientation parameter determined from characteristics of observed information. In a block 102, predictive information and observed information of an object movable in space are received. The object can be a portion of a complex control object such as a hand or other detectable object in a region of space for example. A detectable object is one that is not completely translucent to electromagnetic radiation (including light) at a working wavelength. Common detectable objects useful in various implementations include without limitation a brush, pen or pencil, eraser, stylus, paintbrush and/or other virtualized tool and/or combinations thereof. Objects can be detected, and observed information can be determined, in a variety of ways, but in an implementation and by way of example, one method for detecting objects is described below with reference to flowchart 200 of FIG. 2.

[0074] As shown by flowchart 200 depicted by FIG. 2, in a block 202 sensory information indicating a sensed presence or variance of at least one complex control object is received. Sensory information can be received from a detection system, such as provided by detection system 90 of FIG. 17, recorded image(s) of an imaging system, computer generated virtual image(s), and/or combinations thereof. In an implementation and by way of example, one technique for determining observed information is described below with reference to flowchart 1100 of FIG. 11 and flowcharts 1200 and 1300 of FIGS. 12, 13. Other methods, such as determining point clouds of pixels of like or similar brightness can also be used in some implementations. Continuing with FIG. 2, in a block 212 predictive information defining a complex control object (e.g., a hand, hand portion, tool, tool portion, other body portion or combinations thereof) is received. The predictive information can comprise a model or model sub-component—i.e., one selected to be appropriate to the complex control object—from object library 197A of FIG. 20, for example, or received from another program or source. Predictive information including a model that corresponds to the portion of the hand or other detectable object can be determined in a variety of ways. In an implementation and by way of example, one technique for determining predictive information including a model corresponding to the portion of the hand or other detectable object is described below with reference to flowchart 1400 of FIG. 14 and FIGS. 21, 22. Other modeling techniques (e.g., skeletal models, visual hulls, surface reconstructions, other types of virtual surface or volume reconstruction techniques, or combinations thereof) can be used in other implementations as will be readily apparent to one skilled in the art.

[0075] Again with reference to FIG. 1, in a block 112, an initialization parameter is determined from the observed information. In an implementation and by way of example, one method for determining the initialization parameter is described below with reference to flowchart 300 of FIG. 3 and FIGS. 26-27 and 28. As shown by FIG. 3, a flowchart 300 includes a block 302 in which one or more contours are extracted from a surface portion as represented in the observed information. As illustrated by block 20 of FIG. 26, a surface portion of an imaged object 2610 (e.g., image of real object(s), computer generated input, or combinations thereof) within the observed information can comprise a plurality of surfaces that can be sampled to provide points 2601, from which one or more contours 2602 can be extracted.

[0076] One method for determining a contour portion is illustrated by flowchart 400 of FIG. 4, and block 22 of FIG. 26, includes block 402, in which the observed information can be sampled to determine a set of points. In some implementations, the observed information can comprise an image or images of the imaged object 2610 to be sampled. In other implementations, the observed information comprises instead, a set of points or a set of contours determined from an imaged object 2610. From a set of points 2601, one or more contours 2602 can be determined using any of a variety of techniques, such as for example in block 412, determining a gradient for points determined from the imaged object. In a block 422, a relatively larger value of the gradient is determined. Points along the relatively larger value of the gradient can comprise a contour. As shown by block 22 of FIG. 26, a contour portion A 2602 can be determined from a plurality of points 2601 of block 20.

[0077] Now again with reference to FIG. 3, in a block 304 one or more contour parameters are identified for at least one contour extracted in block 302. As shown by block 22 of FIG. 26 and flowchart 500 depicted in FIG. 5, a contour parameter comprising normal A1 2603 can be determined from contour portion A 2602. Flowchart 500 includes a block 502 in which a first (arbitrary) point P₁ on a contour portion representing a surface portion of interest in the observed information is selected as a starting point. In a block 512, a normal is identified at the point P₁. One method for determining a normal is illustrated by block 22 of FIG. 26 and flowchart 600 of FIG. 6, which includes a block 602 in which a set of points proximate to the first point P₁, at least two of which are not co-linear, is determined. Then, in a block 612, a normal for the first unmatched point can be determined using the other points in the set by determining a normal perpendicular to the plane. For example, given points P₁, P₂, P₃, the normal n is given by the cross product:
n=(p₂−p₁)×(p₃−p₁)

[0078] Another technique that can be used: (i) start with the set of points; (ii) form a first vector from P₂-P₁, (iii) apply rotation matrix to rotate the first vector 90 degrees away from the center of mass of the set of points. (The center of mass of the set of points can be determined by an average of the points). A yet further technique that can be used includes: (i) determine a first vector tangent to a point on a contour in a first image; (ii) determine from the point on the contour a second vector from that point to a virtual camera object in space; (iii) determine a cross product of the first vector and the second vector. The cross product is a normal vector to the contour.

[0079] In implementations, other instances of the contour parameter, i.e., other normal(s), can be determined for other contour portions determined from the sample points 2601 corresponding to the imaged object 2610. Furthermore, instances of different types of contour parameters, e.g., center(s) of mass, (e.g., a weighted average of the points within a set), a curvature(s), and so forth, can be determined for the contour portions determined from the sample points 2601 corresponding to the imaged object 2610.

[0080] Again with reference to FIG. 3, in a block 306, a check whether there are any further parameters to process is made. If there are further parameters to process, then flow continues with block 302 to process the next parameter. Otherwise, in a block 308, at least one contour property is determined for the various contour parameters. One way to determine a contour property illustrated by block 24 of FIG. 26 and flowchart 700 of FIG. 7 includes a block 702 in which one or more contour parameters 2603 are combined to produce a contour property 2604 by applying a function. For example, one or more normal(s) 2603 can be combined to form an average normal 2604. A function appropriate to the implementation is applied to the contour parameters to provide grouping of information from the contour parameters into a contour property. For example, in block 24 of FIG. 26, the average function can be applied to the set of normal(s) determined in block 22 to create an average normal 2604. Of course, in implementations, other functions (mean, mode, variance, interpolation, and so forth and/or combinations thereof depending upon the implementation specific requirements or desirability) can be applied to the contour parameters to arrive at a contour property. In a block 712, the result of applying the function to the one or more contour parameters is provided as the contour property 2604.

[0081] Now again with reference to FIG. 3, in a block 310, an initialization parameter is determined based at least in part upon the at least one contour property determined above with reference to block 308. An initialization parameter can be determined in a variety of ways, such as using one technique illustrated with reference to flowchart 800 of FIG. 8 and block 26 of FIG. 27. Flowchart 800 includes a block 802 in which a plane 2710 (FIG. 27) of best fit is determined through contours in the observed information. A number of techniques are available for determining a plane of best fit illustrated with reference to block 26 of FIG. 27, such as multi-linear regression. In a block 812 a third vector 2714 is determined by combining a contour property 2604 (e.g., average direction of plane normal vectors) with a normal vector 2712 defining the plane 2710 to derive a third vector 2714. One technique employs a cross product to combine contour property 2604 with normal vector 2712, however other techniques can be used in some implementations.

[0082] In a block 822, an initialization parameter 2720 determined from the three vectors determined in block 802 is provided. For example, again with reference to block 26 of FIG. 27, normal vector 2712, contour property 2604 and cross product 2714 can be provided as the initialization parameter 2720. In an implementation, the procedure illustrated in FIG. 3 completes and returns a set of vectors as an initialization parameter. In an alternative implementation, a least one of the vectors (2714 in block 26 of FIG. 27) can be returned as the initialization parameter 2720.

[0083] In a block 122, orientation of the model portion is initialized by applying the initialization parameter to at least a portion of a model within the predictive information. Application of the initialization parameter can be used to initialize orientation of the model portion using a variety of techniques—aligning, weighting, specifying (or computing) confidence factors, selecting model portions, and mode (child's hand, cat's paw, tool tip, and so forth) selection. For example and with reference to block 902 of flowchart 900 of FIG. 9, one or more portions of a model in the predictive information can be aligned using the initialization parameter 2720 as an initial alignment in which to place portion(s) of the object to produce an initialized model. In implementations, aligning can include rotational alignment of the model along one or more axes to correspond to the initialization parameter. In other implementations, the rotational alignment can be applied to portion(s) of the model according to a function F, that determines, based at least in part upon the initialization parameter, that a model portion is too far removed from alignment in one or more directions for example to be used. Function F can exclude (or alter the weighting or confidence factor) for that portion in the model based at least in part upon the initialization parameter. In yet further implementations, the model can be aligned according to the initialization parameter and one or more quantifiable characteristics determined from imaging the object (i.e., brightness, frequencies, and so forth). In a further implementation, with reference to block 906 of flowchart 904 of FIG. 9, a model portion can be selected based at least in part upon a degree to which the model portion(s) align to the initialization parameter. For example, an initialization parameter can be used to select from models for adult hands, children's hands, animal paws, tool tips, and so forth based upon alignment. In yet further implementations, with reference to block 910 of flowchart 908 of FIG. 9, initialization parameters can be used to communicate control information to the MSCS, for example, changing power consumption profiles based on quality of alignment of observed object(s), changing frame rates, selecting user modes in software (tool, hand, face discrimination), background elimination, noise elimination.

[0084] In an optional block 132, a compensation can be applied to a model within the observed. Now with reference to FIG. 10 and FIG. 28, one of a variety of techniques for fitting models to observed information in which one or more elements is missing, either due to differences in the object being observed and/or the viewing environment (e.g., noise, occlusions, poor contrast, and so forth) comprises fitting model portion(s) to extremity portion(s) of an observed object and/or fitting model portion(s) to contour segments and properties. One technique for fitting a model illustrated by block 30 of FIG. 28 and a flowchart 1000 of FIG. 10 in which a block 1002 includes determining observed extremities 2803, 2805 of an object portion(s) 2802 in the observed corresponding to model extremities 2813, 2815 in a default model 2812. In a block 1012, a first quality of fit is determined between the observed information and the default model. In a block 1022, a model compensation technique is applied to the default model 2812 to form a compensated model 2822. In one implementation, the compensation technique is to curve a plurality of model extremities under the model because they are likely occluded in the observed. In a block 1032, a second quality of fit is determined between the observed information and the compensated model. In a block 1042, a determination is made whether the quality of fit to the compensated model is superior to the quality of fit to the default model. If so, the compensated model is adopted in block 1052. Otherwise, in the event that the observed fit the default model better than to the compensated model, processing returns to block 1022 to try another compensation technique if available. In some implementations, more than one compensation can be applied to the model to refine the model for various different occlusions, or the like. Other techniques for determining occlusions for the various model portions that can be used in implementations include best fit analysis of the observed against a set of possible compensated models. Further, techniques can be readily created using variations, combinations, or additions of other techniques to the foregoing teachings within the scope of the technology disclosed.

[0085] In an implementation and by way of example, FIG. 11 illustrates a flowchart 1100 of one technique for detecting objects. Of course, objects can be detected in a variety of ways, and the method of flowchart 1100 is illustrative rather than limiting. In a block 1102, presence or variance of object(s) can be sensed using a detection system 90A (see e.g., FIGS. 17-20 below). In a block 1112, detection system results are analyzed to detect objects, object attributes, motions, and so forth based on changes in image or other sensed parameters (e.g., brightness, and so forth). A variety of analysis methodologies suitable for providing object attribute and/or feature detection based upon sensed parameters can be employed in implementations. Some example analysis implementations are discussed below with reference to FIGS. 12-13. At block 1122, object(s) of interest and background object(s) can be determined using a feature detection algorithms, motion capture algorithms or other methodologies, or combinations thereof. One example of an appropriate feature detection algorithm can be any of the tangent-based algorithms described in co-pending U.S. Ser. No. 13/414,485, filed Mar. 7, 2012, and Ser. No. 13/742,953, filed Jan. 16, 2013; however, other algorithms (e.g., edge detection, axial detection, surface detection techniques, and so forth) can also be used in some implementations.

[0086] Image analysis can be achieved by various algorithms and/or mechanisms. For example, FIG. 12 illustrates a flowchart 1200 of one method for detecting edges or other features of object(s). This analysis implementation can include a block 1202, in which the brightness of two or more pixels is compared to a threshold. In a block 1212, transition(s) in brightness from a low level to a high level across adjacent pixels are detected. In another example, FIG. 13 illustrates a flowchart 1300 of an alternative method for detecting edges or other features of object(s), including a block 1302 of comparing successive images captured with and without illumination by light source(s). In a block 1312, transition(s) in brightness from a low level to a high level across corresponding pixels in the successive images are detected.

[0087] As shown by FIG. 14, a method 1400 of determining predictive information includes a block 1402 in which a model or model sub-component applicable to the object(s) attribute(s) and/or motion(s) based on presence or variance of object(s), sensed using a detection system, such as detection system 90A of FIG. 17, is located in the object library 197A of FIG. 20. Sensing can include capturing image(s), detecting presence with scanning, obtaining other sensory information (e.g., olfactory, pressure, audio or combinations thereof) and/or combinations thereof. In a block 1412, portion(s) of object(s) as detected or captured are analyzed to determine fit to model portion(s) (see e.g., FIGS. 20, 21, 22). In a block 1422, predictive information is refined to include the model portion(s) determined in block 1412. In a block 1432, existence of other sensed object portion(s) is determined. If other object portion(s) have been sensed, then the method continues processing the other object portion(s). Otherwise, each of the sensed object portion(s) has been processed.

[0088] A gesture-recognition system recognizes gestures for purposes of providing input to the electronic device, but can also capture the position and shape of the user's hand in consecutive video images in order to characterize a hand gesture in 3D space and reproduce it on the display screen. A 3D model of the user's hand is determined from a solid hand model covering one or more capsule elements built from the images using techniques described below with reference to FIGS. 15A-15C.

[0089] FIG. 15A shows one implementation of a 3D solid hand model 1500A with capsule representation of predictive information of the hand. Some examples of predictive information of the hand include finger segment length, distance between finger tips, joint angles between fingers, and finger segment orientation. As illustrated by FIG. 15A, the prediction information can be constructed from one or more model subcomponents referred to as capsules 1530, 1532, and 1534, which are selected and/or configured to represent at least a portion of a surface of the hand and virtual surface portion 1522. In some implementations, the model subcomponents can be selected from a set of radial solids, which can reflect at least a portion of the hand in terms of one or more of structure, motion characteristics, conformational characteristics, other types of characteristics of hand, and/or combinations thereof. In one implementation, radial solids are objects made up of a 1D or 2D primitive (e.g., line, curve, plane) and a surface having a constant radial distance to the 1D or 2D primitive. A closest point to the radial solid can be computed relatively quickly. As used herein, three or greater capsules are referred to as a “capsoodle.”

[0090] In an implementation, observation information including observation of the control object can be compared against the model at one or more of periodically, randomly or substantially continuously (i.e., in real time). A “control object” as used herein with reference to an implementation is generally any three-dimensionally movable object or appendage with an associated position and/or orientation (e.g., the orientation of its longest axis) suitable for pointing at a certain location and/or in a certain direction. Control objects include, e.g., hands, fingers, feet, or other anatomical parts, as well as inanimate objects such as pens, styluses, handheld controls, portions thereof, and/or combinations thereof. Where a specific type of control object, such as the user's finger, is used hereinafter for ease of illustration, it is to be understood that, unless otherwise indicated or clear from context, any other type of control object can be used as well.

[0091] Observational information can include without limitation observed values of attributes of the control object corresponding to the attributes of one or more model subcomponents in the predictive information for the control object. In an implementation, comparison of the model with the observation information provides an error indication. In an implementation, an error indication can be computed by determining a closest distance determined between a first point A belonging to a set of points defining the virtual surface 1522 and a second point B belonging to a model subcomponent 1530 determined to be corresponding to the first point (e.g., nearest to the first point for example). In an implementation, the error indication can be applied to the predictive information to correct the model to more closely conform to the observation information. In an implementation, error indication can be applied to the predictive information repeatedly until the error indication falls below a threshold, a measure of conformance with the observation information rises above a threshold, or a fixed or variable number of times, or a fixed or variable number of times per time period, or combinations thereof.

[0092] In one implementation and with reference to FIGS. 15B and 15C, a collection of radial solids and/or capsuloids can be considered a “capsule hand.” In particular, FIGS. 15B and 15C illustrate different views 1500B and 1500C of a 3D capsule hand. A number of capsuloids 1572, e.g. five (5), are used to represent fingers on a hand while a number of radial solids 1574 are used to represent the shapes of the palm and wrist.

[0093] FIGS. 17-20 illustrate an exemplary machine sensory and control system (MSCS) in implementations.

[0094] In one implementation, a motion sensing and controller system provides for detecting that some variation(s) in one or more portions of interest of a user has occurred, for determining that an interaction with one or more machines corresponds to the variation(s), for determining if the interaction should occur, and, if so, for affecting the interaction. The Machine Sensory and Control System (MSCS) typically includes a portion detection system, a variation determination system, an interaction system and an application control system.

[0095] As FIG. 17 shows, one detection system 90A implementation includes an emission module 91, a detection module 92, a controller 96, a processing module 94 and a machine control module 95. In one implementation, the emission module includes one or more emitter(s) 180A, 180B (e.g., LEDs or other devices emitting light in the IR, visible, or other spectrum regions, or combinations thereof; radio and/or other electromagnetic signal emitting devices) that are controllable via emitter parameters (e.g., frequency, activation state, firing sequences and/or patterns, etc.) by the controller 96. However, other existing/emerging emission mechanisms and/or some combination thereof can also be utilized in accordance with the requirements of a particular implementation. The emitters 180A, 180B can be individual elements coupled with materials or devices 182 (and/or materials) (e.g., lenses 182A, multi-lenses 182B (of FIG. 21), image directing film (IDF) 182C (of FIG. 18), liquid lenses, combinations thereof, and/or others) with varying or variable optical properties to direct the emission, one or more arrays 180C of emissive elements (combined on a die or otherwise), with or without the addition of devices 182C for directing the emission, or combinations thereof, and positioned within an emission region 181 (of FIG. 18) according to one or more emitter parameters (i.e., either statically (e.g., fixed, parallel, orthogonal or forming other angles with a work surface, one another or a display or other presentation mechanism) or dynamically (e.g., pivot, rotate and/or translate) mounted, embedded (e.g., within a machine or machinery under control) or otherwise coupleable using an interface (e.g., wired or wireless)). In some implementations, structured lighting techniques can provide improved surface feature capture capability by casting illumination according to a reference pattern onto the object 98. Image capture techniques described in further detail herein can be applied to capture and analyze differences in the reference pattern and the pattern as reflected by the object 98. In yet further implementations, detection system 90A may omit emission module 91 altogether (e.g., in favor of ambient lighting).

[0096] In one implementation, the detection module 92 includes one or more capture device(s) 190A, 190B (e.g., light (or other electromagnetic radiation sensitive devices) that are controllable via the controller 96. The capture device(s) 190A, 190B can comprise individual or multiple arrays of image capture elements 190A (e.g., pixel arrays, CMOS or CCD photo sensor arrays, or other imaging arrays) or individual or arrays of photosensitive elements 190B (e.g., photodiodes, photo sensors, single detector arrays, multi-detector arrays, or other configurations of photo sensitive elements) or combinations thereof. Arrays of image capture device(s) 190C (of FIG. 19) can be interleaved by row (or column or a pattern or otherwise addressable singly or in groups). However, other existing/emerging detection mechanisms and/or some combination thereof can also be utilized in accordance with the requirements of a particular implementation. Capture device(s) 190A, 190B each can include a particular vantage point 190-1 from which objects 98 within area of interest 5 are sensed and can be positioned within a detection region 191 (of FIG. 19) according to one or more detector parameters (i.e., either statically (e.g., fixed, parallel, orthogonal or forming other angles with a work surface, one another or a display or other presentation mechanism) or dynamically (e.g. pivot, rotate and/or translate), mounted, embedded (e.g., within a machine or machinery under control) or otherwise coupleable using an interface (e.g., wired or wireless)). Capture devices 190A, 190B can be coupled with devices 192 (and/or materials) (of FIG. 19) (e.g., lenses 192A (of FIG. 19), multi-lenses 192B (of FIG. 19), image directing film (IDF) 192C (of FIG. 19), liquid lenses, combinations thereof, and/or others) with varying or variable optical properties for directing the reflectance to the capture device for controlling or adjusting resolution, sensitivity and/or contrast. Capture devices 190A, 190B can be designed or adapted to operate in the IR, visible, or other spectrum regions, or combinations thereof; or alternatively operable in conjunction with radio and/or other electromagnetic signal emitting devices in various applications. In an implementation, capture devices 190A, 190B can capture one or more images for sensing objects 98 and capturing information about the object (e.g., position, motion, etc.). In implementations comprising more than one capture device, particular vantage points of capture devices 190A, 190B can be directed to area of interest 5 so that fields of view 190-2 of the capture devices at least partially overlap. Overlap in the fields of view 190-2 provides capability to employ stereoscopic vision techniques (see, e.g., FIG. 19), including those known in the art to obtain information from a plurality of images captured substantially contemporaneously.

[0097] While illustrated with reference to a particular implementation in which control of emission module 91 and detection module 92 are co-located within a common controller 96, it should be understood that these functions will be separate in some implementations, and/or incorporated into one or a plurality of elements comprising emission module 91 and/or detection module 92 in some implementations. Controller 96 comprises control logic (hardware, software or combinations thereof) to conduct selective activation/de-activation of emitter(s) 180A, 180B (and/or control of active directing devices) in on-off, or other activation states or combinations thereof to produce emissions of varying intensities in accordance with a scan pattern which can be directed to scan an area of interest 5. Controller 96 can comprise control logic (hardware, software or combinations thereof) to conduct selection, activation and control of capture device(s) 190A, 190B (and/or control of active directing devices) to capture images or otherwise sense differences in reflectance or other illumination. Signal processing module 94 determines whether captured images and/or sensed differences in reflectance and/or other sensor—perceptible phenomena indicate a possible presence of one or more objects of interest 98, including control objects 99, the presence and/or variations thereof can be used to control machines and/or other applications 95.

[0098] In various implementations, the variation of one or more portions of interest of a user can correspond to a variation of one or more attributes (position, motion, appearance, surface patterns) of a user hand 99, finger(s), points of interest on the hand 99, facial portion 98 other control objects (e.g., styli, tools) and so on (or some combination thereof) that is detectable by, or directed at, but otherwise occurs independently of the operation of the machine sensory and control system. Thus, for example, the system is configurable to ‘observe’ ordinary user locomotion (e.g., motion, translation, expression, flexing, deformation, and so on), locomotion directed at controlling one or more machines (e.g., gesturing, intentionally system-directed facial contortion, etc.), attributes thereof (e.g., rigidity, deformation, fingerprints, veins, pulse rates and/or other biometric parameters). In one implementation, the system provides for detecting that some variation(s) in one or more portions of interest (e.g., fingers, fingertips, or other control surface portions) of a user has occurred, for determining that an interaction with one or more machines corresponds to the variation(s), for determining if the interaction should occur, and, if so, for at least one of initiating, conducting, continuing, discontinuing and/or modifying the interaction and/or a corresponding interaction.

[0099] For example and with reference to FIG. 20, a variation determination system 90B implementation comprises a model management module 197 that provides functionality to build, modify, customize one or more models to recognize variations in objects, positions, motions and attribute state and/or change in attribute state (of one or more attributes) from sensory information obtained from detection system 90A. A motion capture and sensory analyzer 197E finds motions (i.e., translational, rotational), conformations, and presence of objects within sensory information provided by detection system 90A. The findings of motion capture and sensory analyzer 197E serve as input of sensed (e.g., observed) information from the environment with which model refiner 197F can update predictive information (e.g., models, model portions, model attributes, etc.).

[0100] A model management module 197 implementation comprises a model refiner 197F to update one or more models 197B (or portions thereof) from sensory information (e.g., images, scans, other sensory-perceptible phenomenon) and environmental information (i.e., context, noise, etc.); enabling a model analyzer 197I to recognize object, position, motion and attribute information that might be useful in controlling a machine. Model refiner 197F employs an object library 197A to manage objects including one or more models 197B (i.e., of user portions (e.g., hand, face), other control objects (e.g., styli, tools)) or the like (see e.g., model 197B-1, 197B-2 of FIGS. 23A, 23B)), model components (i.e., shapes, 2D model portions that sum to 3D, outlines 194 and/or outline portions 194A, 194B (i.e., closed curves), attributes 197-5 (e.g., attach points, neighbors, sizes (e.g., length, width, depth), rigidity/flexibility, torsional rotation, degrees of freedom of motion and others) and so forth) (see e.g., 197B-1-197B-2 of FIGS. 23A-23B), useful to define and update models 197B, and model attributes 197-5. While illustrated with reference to a particular implementation in which models, model components and attributes are co-located within a common object library 197A, it should be understood that these objects will be maintained separately in some implementations.

[0101] In an implementation, when the control object morphs, conforms, and/or translates, motion information reflecting such motion(s) is included into the observed information. Points in space can be recomputed based on the new observation information. The model subcomponents can be scaled, sized, selected, rotated, translated, moved, or otherwise re-ordered to enable portions of the model corresponding to the virtual surface(s) to conform within the set of points in space.

[0102] In an implementation, motion(s) of the control object can be rigid transformation, in which case, points on the virtual surface(s) remain at the same distance(s) from one another through the motion. Motion(s) can be non-rigid transformations, in which points on the virtual surface(s) can vary in distance(s) from one another during the motion. In an implementation, observation information can be used to adjust (and/or recomputed) predictive information thereby enabling “tracking” the control object. In implementations, control object can be tracked by determining whether a rigid transformation or a non-rigid transformation occurs. In an implementation, when a rigid transformation occurs, a transformation matrix is applied to each point of the model uniformly. Otherwise, when a non-rigid transformation occurs, an error indication can be determined, and an error minimization technique such as described herein above can be applied. In an implementation, rigid transformations and/or non-rigid transformations can be composed. One example composition implementation includes applying a rigid transformation to predictive information. Then an error indication can be determined, and an error minimization technique such as described herein above can be applied. In an implementation, determining a transformation can include calculating a rotation matrix that provides a reduced RMSD (root mean squared deviation) between two paired sets of points. One implementation can include using Kabsch Algorithm to produce a rotation matrix. In an implementation and by way of example, one or more force lines can be determined from one or more portions of a virtual surface.

[0103] FIG. 21 illustrates prediction information including a model 197B-1 of a control object (e.g., FIG. 17: 99) constructed from one or more model subcomponents 197-2, 197-3 selected and/or configured to represent at least a portion of a surface of control object 99, a virtual surface portion 194 and one or more attributes 197-5. Other components can be included in prediction information 197B-1 not shown in FIG. 21 for clarity sake. In an implementation, the model subcomponents 197-2, 197-3 can be selected from a set of radial solids, which can reflect at least a portion of a control object 99 in terms of one or more of structure, motion characteristics, conformational characteristics, other types of characteristics of control object 99, and/or combinations thereof. In one implementation, radial solids include a contour and a surface defined by a set of points having a fixed distance from the closest corresponding point on the contour. Another radial solid implementation includes a set of points normal to points on a contour and a fixed distance therefrom. In an implementation, computational technique(s) for defining the radial solid include finding a closest point on the contour and the arbitrary point, then projecting outward the length of the radius of the solid. In an implementation, such projection can be a vector normal to the contour at the closest point. An example radial solid (e.g., 197-3) includes a “capsuloid”, i.e., a capsule shaped solid including a cylindrical body and semi-spherical ends. Another type of radial solid (e.g., 197-2) includes a sphere. Other types of radial solids can be identified based on the foregoing teachings.

[0104] One or more attributes 197-5 can define characteristics of a model subcomponent 197-3. Attributes can include e.g., attach points, neighbors, sizes (e.g., length, width, depth), rigidity, flexibility, torsion, zero or more degrees of freedom of motion with respect to one or more defined points, which can include endpoints for example, and other attributes defining a salient characteristic or property of a portion of control object 99 being modeled by predictive information 197B-1. In an implementation, predictive information about the control object can include a model of the control object together with attributes defining the model and values of those attributes.

[0105] In an implementation, observation information including observation of the control object can be compared against the model at one or more of periodically, randomly or substantially continuously (i.e., in real time). Observational information can include without limitation observed values of attributes of the control object corresponding to the attributes of one or more model subcomponents in the predictive information for the control object. In an implementation, comparison of the model with the observation information provides an error indication. In an implementation, an error indication can be computed by determining a closest distance determined between a first point A belonging to a set of points defining the virtual surface 194 and a second point B belonging to a model subcomponent 197-2 determined to be corresponding to the first point (e.g., nearest to the first point for example). In an implementation, the error indication can be applied to the predictive information to correct the model to more closely conform to the observation information. In an implementation, error indication can be applied to the predictive information repeatedly until the error indication falls below a threshold, a measure of conformance with the observation information rises above a threshold, or a fixed or variable number of times, or a fixed or variable number of times per time period, or combinations thereof.

[0106] In an implementation and with reference to FIGS. 17, 22, updating predictive information to observed information comprises selecting one or more sets of points (e.g., FIG. 22:193A, 193B) in space surrounding or bounding the control object within a field of view of one or more image capture device(s). As shown by FIG. 22, points 193 can be determined using one or more sets of lines 195A, 195B, 195C, and 195D originating at vantage point(s) (e.g., FIG. 17: 190-1, 190-2) associated with the image capture device(s) (e.g., FIG. 17: 190A-1, 190A-2) and determining therefrom one or more intersection point(s) defining a bounding region (i.e., region formed by lines FIG. 22: 195A, 195B, 195C, and 195D) surrounding a cross-section of the control object. The bounding region can be used to define a virtual surface (FIG. 22: 194) to which model subcomponents 197-1, 197-2, 197-3, and 197-4 can be compared. The virtual surface 194 can include a visible portion 194A and a non-visible “inferred” portion 194B. Virtual surfaces 194 can include straight portions and/or curved surface portions of one or more virtual solids (i.e., model portions) determined by model refiner 197F.

[0107] For example and according to one implementation illustrated by FIG. 22, model refiner 197F determines to model subcomponent 197-1 of an object portion (happens to be a finger) using a virtual solid, an ellipse in this illustration, or any of a variety of 3D shapes (e.g., ellipsoid, sphere, or custom shape) and/or 2D slice(s) that are added together to form a 3D volume. Accordingly, beginning with generalized equations for an ellipse (1) with (x, y) being the coordinates of a point on the ellipse, (x_C, y_C) the center, a and b the axes, and θ the rotation angle. The coefficients C₁, C₂ and C₃ are defined in terms of these parameters, as shown:

[0108] $\begin{array}{cc}{C}_1{x}^2+C_2\mathrm{xy}+C_3{y}^2-\left(2C_1{x}_c+C_2{y}_c\right)x-\left(2C_3{y}_c+C_2{x}_c\right)y+\left(C_1{x}_c^2+C_2{x}_c{y}_c+C_3{y}_c^2-1\right)=0C_1=\frac{{\cos }^2\theta }{a^2}+\frac{{\sin }^2\theta }{b^2}{C}_2=-2\cos \mathrm{\theta sin}\theta \left(\frac{1}{a^2}-\frac{1}{b^2}\right)C_3=\frac{{\sin }^2\theta }{a^2}+\frac{{\cos }^2\theta }{b^2}& \left(1\right)\end{array}$

[0109] The ellipse equation (1) is solved for θ, subject to the constraints that: (1) (x_C, y_C) must lie on the centerline determined from the four tangents 195A, 195B, 195C, and 195D (i.e., centerline 189A of FIG. 22); and (2) a is fixed at the assumed value a₀. The ellipse equation can either be solved for θ analytically or solved using an iterative numerical solver (e.g., a Newtonian solver as is known in the art). An analytic solution can be obtained by writing an equation for the distances to the four tangent lines given a y_C position, then solving for the value of y_C that corresponds to the desired radius parameter a=a₀. Accordingly, equations (2) for four tangent lines in the x-y plane (of the slice), in which coefficients A_i, B_i and D_i (for i=1 to 4) are determined from the tangent lines 195A, 195B, 195C, and 195D identified in an image slice as described above.
A₁x+B₁y+D₁=0
A₂x+B₂y+D₂=0
A₃x+B₃y+D₃=0
A₄x+B₄y+D₄=0  (2)

[0110] Four column vectors r₁₂, r₂₃, r₁₄ and r₂₄ are obtained from the coefficients A_i, B_i and D_i of equations (2) according to equations (3), in which the “\” operator denotes matrix left division, which is defined for a square matrix M and a column vector v such that M\v=r, where r is the column vector that satisfies Mr=v:

[0111] $\begin{array}{cc}{r}_{13}=\left[\begin{array}{cc}{A}_1& B_1\\ A_3& B_3\end{array}\right]\backslash \left[\begin{array}{c}-D_1\\ -D_3\end{array}\right]r_{23}=\left[\begin{array}{cc}{A}_2& B_2\\ A_3& B_3\end{array}\right]\backslash \left[\begin{array}{c}-D_{21}\\ -D_3\end{array}\right]r_{14}=\left[\begin{array}{cc}{A}_1& B_1\\ A_4& B_4\end{array}\right]\backslash \left[\begin{array}{c}-D_1\\ -D_4\end{array}\right]r_{24}=\left[\begin{array}{cc}{A}_2& B_2\\ A_4& B_4\end{array}\right]\backslash \left[\begin{array}{c}-D_2\\ -D_4\end{array}\right]& \left(3\right)\end{array}$

[0112] Four component vectors G and H are defined in equations (4) from the vectors of tangent coefficients A, B and D and scalar quantities p and q, which are defined using the column vectors r₁₂, r₂₃, r₁₄ and r₂₄ from equations (3).
c1=(r₁₃+r₂₄)/2
c2=(r₁₄+r₂₃)/2
δ1=c2₁−c1₁
δ2=c2₂−c1₂
p=δ1/δ2
q=c1₁−c1₂*P
G=Ap+B
H=Aq+D  (4)

[0113] Six scalar quantities v_A2, v_AB, v_B2, w_A2, w_AB, and w_B2 are defined by equation (5) in terms of the components of vectors G and H of equation (4).

[0114] $\begin{array}{cc}v=\left[\begin{array}{ccc}{G}_2^2& G_3^2& G_4^2\\ {\left(G_2{H}_2\right)}^2& {\left(G_3{H}_3\right)}^2& {\left(G_4{H}_4\right)}^2\\ H_2^2& H_3^2& H_4^2\end{array}\right]\ddots \left[\begin{array}{c}0\\ 0\\ 1\end{array}\right]w=\left[\begin{array}{ccc}{G}_2^2& G_3^2& G_4^2\\ {\left(G_2{H}_2\right)}^2& {\left(G_3{H}_3\right)}^2& {\left(G_4{H}_4\right)}^2\\ H_2^2& H_3^2& H_4^2\end{array}\right]\ddots \left[\begin{array}{c}0\\ 1\\ 0\end{array}\right]& \left(5\right)\end{array}$
v_A2=(v₁A₁)²+(v₂A₂)²+(v₃A₃)²
v_AB=(v₁A₁B₁)²+(v₂A₂B₂)²+(v₃A₃B₃)²
v_B2=(v₁B₁)²+(v₂B₂)²+(v₃B₃)²
w_A2=(w₁A₁)²+(w₂A₂)²+(w₃A₃)²
w_AB=(w₁A₁B₁)²+(w₂A₂B₂)²+(w₃A₃B₃)²
w_B2=(w₁B₁)²+(w₂B₂)²+(w₃B₃)²

[0115] Using the parameters defined in equations (1)-(5), solving for θ is accomplished by solving the eighth-degree polynomial equation (6) for t, where the coefficients Q_i (for i=0 to 8) are defined as shown in equations (7)-(15).
0=Q₈t⁸+Q₇t⁷+Q₆t⁶+Q₅t⁵+Q₄t⁴+Q₃t³+Q₂t²+Q₁t+Q₀  (6)

[0116] The parameters A₁, B₁, G₁, H₁, v_A2, v_AB, v_B2, w_A2, w_AB, and w_B2 used in equations (7)-(15) are defined as shown in equations (1)-(4). The parameter n is the assumed semi-major axis (in other words, a₀). Once the real roots t are known, the possible values of θ are defined as θ=atan(t).
Q₈=4A²₁n²v_B2²+4v_B2B₁²(1−n²v_A2)
−(G₁(1−n²v_A2)w_B2+n²v_B2w_A2+2H₁v_B2)²  (7)
Q₇=−(2(2n²v_ABw_A2+4H₁v_AB+2G₁n²v_ABw_B2
+2G₁(1−n²v_A2)w_AB))(G₁(1−n²v_A2)w_B2+n²v_B2w_A2+2H₁v_B2)
−8A₁B₁n²v_B2²+16A₁²v_ABv_B2
+(4(2A₁B₁(1−n²v_A2)+2B₁²n²v_AB))v_B2+8B₁²(1−n²v_A2)v_AB  (8)
Q₆=−(2(2H₁v_B2+2H₁v_A2+n²v_A2w_A2+n²v_B2(−2w_AB+w_B2)
+G₁(n²v_B2+1)w_B2+4G₁n²v_ABw_AB+G₁(1−n²v_A2)v_A2))×(G₁(1−n²v_A2)w_B2+n²v_B2w_A2+2H₁v_B2)
−(2n²v_ABw_A2+4H₁v_AB+2G₁n²v_ABw_B2+2G₁(1−n²v_A2)w_AB²
+4B₁²n²v_B2²−32A₁B₁n²v_ABv_B2+4A₁²n²(2v_A2v_B2+4v_AB²)+4A₁²n²v_B2²
+(4(A₁²(1−n²v_A2)+4A₁B₁n²v_AB+B₁²(−n²v_B2+1)
+B₁²(1−n²v_A2)))v_B2+(8(2A₁B₁(1−n²v_A2)+2B₁²n²v_AB))v_AB
+4B₁²(1−n²v_A2)v_A2  (9)
Q₅=−(2(4H₁v_AB+2G₁(−n²v_B2+1)w_AB+2G₁n²v_ABv_A2
+2n²v_A(−2w_AB+w_B2)))(G₁(1−n²v_A2)w_B2+n²v_B2w_A2+2H₁v_B2)
−(2(2H₁v_B2+2H₁v_A2+n²v_A2w_A2+n²v_B2(−2w_AB+w_B2)
+G₁(−n²v_B2+1)w_B2+4G₁n²v_ABw_AB
+G₁(1−n²v_A2)v_A2))×(2n₂v_ABw_A2+4H₁v_AB+2G₁n²v_ABw_B2
+2G₁(1−n²v_A2)w_AB)+16B₁²n²v_ABv_B2
−8A₁B₁²(2v_A2v_B2+4v_AB²)+16A₁²n²v_A2v_AB−8A₁B₁n²v_B2²
+16A₁²n²v_ABv_B2
+(4(2A₁²n²v_AB+2A₁B₁(−n²v_B2+1)+2A₁B₁(1−n²v_A2)+2B₁²n²v_AB))v_B2
+(8(A₁²(1−n²v_A2)+4A₁B₁n²v_AB+B₁²(−n²v_B2+1)
+B₁²(1−n²v_A2)))v_AB+(4(2A₁B₁(1−n²v_A2)+2B₁²n²v_AB  (10)
Q₄=(4(A₁²(−n²v_B2+A₁²(1−n²v_A2)+4A₁B₁n²v_AB+B₁²(−n²v_B2+1)))v_B2
+(8(2A₁²n²v_AB+2A₁B₁(−n²v_B2+1)+2A₁B₁(1−n²v_A2)+2B₁²n²v_AB))v_AB
+(4(A₁²(1−n²v_A2)+4A₁B₁n²v_AB+B₁²(−n²v_B2+1)
+B₁²(1−n²v_A2)))v_A2+4B₁²n²(2v_A2v_B2+4v_AB²)−32A₁B₁n²v_A2v_AB
+4A₁²n²v_A2²+4B₁²n²v_B2²−32A₁B₁n²v_ABv_B2
+4A₁²n²(2v_A2v_B2+4v_AB²)
−(2(G₁(−n²v_B2+1)v_A2+n²v_A2(−2w_AB+w_B2)
+2H₁v_A2))(G₁(1−n²v_A2)w_B2+n²v_B2w_A2+2H₁v_B2)
−(2(4H₁v_AB+2G₁(−n²v_B2+1)w_AB+2G₁n²v_ABv_A2
+2n²v_AB(−2w_AB+w_B2)))×(2n²v_ABw_A2+4H₁v_AB+2G₁n²v_ABw_B2
+2G₁(1−n²v_A2)w_AB)
−(2H₁v_B2+2H₁v_A2+n²v_A2w_A2+n²v_B2(−2w_AB+w_B2)
+G₁(−n²v_B2+1)w_B2+4G₁n²v_ABw_AB+G₁(1−n²v_A2)v_A2)²  (11)
Q₃=(2(G₁(−n²v_B2+1)v_A2+n²v_A2(−2w_AB+w_B2)+2H₁v_A2))(2n²v_ABw_A2
+4H₁v_AB+2G₁n²v_ABw_B2+G₁(1−n²v_A2)w_AB)
−(2(4H₁v_AB+2G₁(−n²v_B2+1)w_AB+2G₁n²v_ABv_A2
+2n²v_AB(−2w_AB+w_B2)))×(2H₁v_B2+2H₁v_A2+n²v_A2w_A2
+n²v_B2(−2w_AB+w_B2)+G₁(−n²v_B2+1)w_B2+4G₁n²v_ABw_AB
+G₁(1−n²v_A2)v_A2)+16B₁²n²v_A2v_AB−8A₁B₁n²v_A2²v_A2²+16B₁²n²v_ABw_B2
−8A₁B₁n²(2v_A2v_B2+4v_AB²)+16A₁²n²v_A2v_AB
+(4(2A₁²n²v_AB+2A₁B₁(−n²v_B2+1)))v_B2
+(8(A₁²(−n²v_B2+1)+A₁²(1−n²v_A2)+4A₁B₁n²v_AB
+B₁²(−n²v_B2+1)))v_AB
+(4(2A₁²n²v_AB+2A₁B₁(−n²v_B2+1)+2A₁B₁(1−n²v_A2)+2B₁²n²v_AB))v_A2  (12)
Q₂=4A₁²(−n²v_B2+1)v_B2+(8(2A₁²n²v_AB+2A₁B₁(−n²v_B2+1)))v_AB
+(4(A₁²(−n²v_B2+1)+A₁²(1−n²v_A2)+4A₁B₁n²v_AB
+B₁²(−n²v_B2+1)))v_A2+4B₁²n²v_A2²+4B₁²n²(2v_A2v_B2+4v_AB²)
−32A₁B₁n²v_A2v_AB+4A₁²n²v_A2²
−(2(G₁(−n²v_B2+1)v_A2+n²v_A2(−2w_AB+w_B2)+2H₁v_A2))×(2H₁v_B2+2H₁v_A2+n²v_A2w_A2+n²v_B2(−2w_AB+w_B2)
+G₁(−n²v_B2+1)w_B2+4G₁n²v_ABw_AB+G₁(1−n²v_A2)v_A2)
−(4H₁v_AB+2G₁(−n²v_B2+1)w_AB+2G₁n²v_ABv_A2
+2n²v_AB(−2w_AB+w_B2))²  (13)
Q₁=8A₁²(−n²v_B2+1)v_AB+(4(2A₁²n²v_AB+2A₁B₁(−n²v_B2+1)))v_A2
+16B₁²n²v_A2v_AB−8A₁B₁n²v_A2²
−(2(G₁(−n²v_B2+1)v_A2+n²v_A2(−2w_AB+w_B2)+2H₁v_A2))(4H₁v_AB
+2G₁(−n²v_B2+1)w_AB+²G₁n²v_ABv_A2+2n²v_AB(−2w_AB+w_B2))  (14)
Q₀=4A₁²(−n²v_B2+1)v_A2−(G₁(−n²v_B2+1)v_A2+n²v_A2(−2w_AB+w_B2)+2H₁v_A2)²  (15)

[0117] In this exemplary implementation, equations (6)-(15) have at most three real roots; thus, for any four tangent lines, there are at most three possible ellipses that are tangent to all four lines and that satisfy the a=a₀ constraint. (In some instances, there may be fewer than three real roots.) For each real root θ, the corresponding values of (x_C, y_C) and b can be readily determined. Depending on the particular inputs, zero or more solutions will be obtained; for example, in some instances, three solutions can be obtained for a typical configuration of tangents. Each solution is completely characterized by the parameters {θ, a=a₀, b, (x_C, y_C)}. Alternatively, or additionally, a model builder 197C and model updater 197D provide functionality to define, build and/or customize model(s) 197B using one or more components in object library 197A. Once built, model refiner 197F updates and refines the model, bringing the predictive information of the model in line with observed information from the detection system 90A.

[0118] The model subcomponents 197-1, 197-2, 197-3, and 197-4 can be scaled, sized, selected, rotated, translated, moved, or otherwise re-ordered to enable portions of the model corresponding to the virtual surface(s) to conform within the points 193 in space. Model refiner 197F employs a variation detector 197G to substantially continuously determine differences between sensed information and predictive information and provide to model refiner 197F a variance useful to adjust the model 197B accordingly. Variation detector 197G and model refiner 197F are further enabled to correlate among model portions to preserve continuity with characteristic information of a corresponding object being modeled, continuity in motion, and/or continuity in deformation, conformation and/or torsional rotations.

[0119] An environmental filter 197H reduces extraneous noise in sensed information received from the detection system 90A using environmental information to eliminate extraneous elements from the sensory information. Environmental filter 197H employs contrast enhancement, subtraction of a difference image from an image, software filtering, and background subtraction (using background information provided by objects of interest determiner 198H (see below) to enable model refiner 197F to build, refine, manage and maintain model(s) 197B of objects of interest from which control inputs can be determined.

[0120] A model analyzer 197I determines that a reconstructed shape of a sensed object portion matches an object model in an object library; and interprets the reconstructed shape (and/or variations thereon) as user input. Model analyzer 197I provides output in the form of object, position, motion and attribute information to an interaction system 90C.

[0121] Again with reference to FIG. 20, an interaction system 90C includes an interaction interpretation module 198 that provides functionality to recognize command and other information from object, position, motion and attribute information obtained from variation system 90B. An interaction interpretation module 198 implementation comprises a recognition engine 198F to recognize command information such as command inputs (i.e., gestures and/or other command inputs (e.g., speech, etc.)), related information (i.e., biometrics), environmental information (i.e., context, noise, etc.) and other information discernable from the object, position, motion and attribute information that might be useful in controlling a machine. Recognition engine 198F employs gesture properties 198A (e.g., path, velocity, acceleration, etc.), control objects determined from the object, position, motion and attribute information by an objects of interest determiner 198H and optionally one or more virtual constructs 198B (see e.g., FIGS. 8-3, 8-4: 198B-1, 198B-2) to recognize variations in control object presence or motion indicating command information, related information, environmental information and other information discernable from the object, position, motion and attribute information that might be useful in controlling a machine. With reference to FIG. 20-3, 8-4, virtual construct 198B-1, 198B-2 implement an engagement target with which a control object 99 interacts—enabling MSCS 189 to discern variations in control object (i.e., motions into, out of or relative to virtual construct 198B) as indicating control or other useful information. A gesture trainer 198C and gesture properties extractor 198D provide functionality to define, build and/or customize gesture properties 198A.

[0122] A context determiner 198G and object of interest determiner 198H provide functionality to determine from the object, position, motion and attribute information objects of interest (e.g., control objects, or other objects to be modeled and analyzed), objects not of interest (e.g., background) based upon a detected context. For example, when the context is determined to be an identification context, a human face will be determined to be an object of interest to the system and will be determined to be a control object. On the other hand, when the context is determined to be a fingertip control context, the finger tips will be determined to be object(s) of interest and will be determined to be a control objects whereas the user's face will be determined not to be an object of interest (i.e., background). Further, when the context is determined to be a styli (or other tool) held in the fingers of the user, the tool tip will be determined to be object of interest and a control object whereas the user's fingertips might be determined not to be objects of interest (i.e., background). Background objects can be included in the environmental information provided to environmental filter 197H of model management module 197.

[0123] A virtual environment manager 198E provides creation, selection, modification and de-selection of one or more virtual constructs 198B (see FIGS. 8-3, 8-4). In some implementations, virtual constructs (e.g., a virtual object defined in space; such that variations in real objects relative to the virtual construct, when detected, can be interpreted for control or other purposes (see FIGS. 8-3, 8-4)) are used to determine variations (i.e., virtual “contact” with the virtual construct, breaking of virtual contact, motion relative to a construct portion, etc.) to be interpreted as engagements, dis-engagements, motions relative to the construct(s), and so forth, enabling the system to interpret pinches, pokes and grabs, and so forth. Interaction interpretation module 198 provides as output the command information, related information and other information discernable from the object, position, motion and attribute information that might be useful in controlling a machine from recognition engine 198F to an application control system 90D.

Collisions

[0124] In an implementation, predictive information can include collision information concerning two or more capsoloids. By means of illustration, several possible fits of predicted information to observed information can be removed from consideration based upon a determination that these potential solutions would result in collisions of capsoloids. In an implementation, a relationship between neighboring capsoloids, each having one or more attributes (e.g., determined minima and/or maxima of intersection angles between capsoloids) can be determined. In an implementation, determining a relationship between a first capsoloid having a first set of attributes and a second capsoloid having a second set of attributes includes detecting and resolving conflicts between first attribute and second attributes. For example, a conflict can include a capsoloid having one type of angle value with a neighbor having a second type of angle value incompatible with the first type of angle value. Attempts to attach a capsoloid with a neighboring capsoloid having attributes such that the combination will exceed what is allowed in the observed—or to pair incompatible angles, lengths, shapes, or other such attributes—can be removed from the predicted information without further consideration.

Lean Model

[0125] In an implementation, predictive information can be artificially constrained to capsoloids positioned in a subset of the observed information—thereby enabling creation of a “lean model”. For example, as illustrated in FIG. 21, capsoloid 197-3 could be used to denote the portion of the observed without addition of capsoloids 197-2. In a yet further implementation, connections can be made using artificial constructs to link together capsoloids of a lean model. In another implementation, the predictive information can be constrained to a subset of topological information about the observed information representing the control object to form a lean model.

[0126] In an implementation, a lean model can be associated with a full predictive model. The lean model (or topological information, or properties described above) can be extracted from the predictive model to form a constraint. Then, the constraint can be imposed on the predictive information thereby enabling the predictive information to be constrained in one or more of behavior, shape, total (system) energy, structure, orientation, compression, shear, torsion, other properties, and/or combinations thereof.

Occlusions

[0127] In an implementation, the observed can include components reflecting portions of the control object which are occluded from view of the device (“occlusions” or “occluded components”). In one implementation, the predictive information can be “fit” to the observed as described herein above with the additional constraint(s) that some total property of the predictive information (e.g., potential energy) be minimized or maximized (or driven to lower or higher value(s) through iteration or solution). Properties can be derived from nature, properties of the control object being viewed, others, and/or combinations thereof. In another implementation, as shown by FIGS. 16A and 16B, a deformation of the predictive information subcomponents 1602 and 1612 can be allowed subject to an overall permitted value of compression, deformation, flexibility, others, and/or combinations thereof.

Friction

[0128] In an implementation, a “friction constraint” is applied on the model 197B-1. For example, if fingers of a hand being modeled are close together (in position or orientation), corresponding portions of the model will have more “friction”. The more friction a model subcomponent has in the model, the less the subcomponent moves in response to new observed information. Accordingly the model is enabled to mimic the way portions of the hand that are physically close together move together, and move less overall.

[0129] An environmental filter 197H reduces extraneous noise in sensed information received from the detection system 90A using environmental information to eliminate extraneous elements from the sensory information. Environmental filter 197H employs contrast enhancement, subtraction of a difference image from an image, software filtering, and background subtraction (using background information provided by objects of interest determiner 198H (see below) to enable model refiner 197F to build, refine, manage and maintain model(s) 197B of objects of interest from which control inputs can be determined.

[0130] A model analyzer 197I determines that a reconstructed shape of a sensed object portion matches an object model in an object library; and interprets the reconstructed shape (and/or variations thereon) as user input. Model analyzer 197I provides output in the form of object, position, motion and attribute information to an interaction system 90C.

[0131] Again with reference to FIG. 20, an interaction system 90C includes an interaction interpretation module 198 that provides functionality to recognize command and other information from object, position, motion and attribute information obtained from variation system 90B. An interaction interpretation module 198 implementation comprises a recognition engine 198F to recognize command information such as command inputs (i.e., gestures and/or other command inputs (e.g., speech, etc.)), related information (i.e., biometrics), environmental information (i.e., context, noise, etc.) and other information discernable from the object, position, motion and attribute information that might be useful in controlling a machine. Recognition engine 198F employs gesture properties 198A (e.g., path, velocity, acceleration, etc.), control objects determined from the object, position, motion and attribute information by an objects of interest determiner 198H and optionally one or more virtual constructs 198B (see e.g., FIGS. 8-3, 8-4: 198B-1, 198B-2) to recognize variations in control object presence or motion indicating command information, related information, environmental information and other information discernable from the object, position, motion and attribute information that might be useful in controlling a machine. With reference to FIG. 20-3, 8-4, virtual construct 198B-1, 198B-2 implement an engagement target with which a control object 99 interacts—enabling MSCS 189 to discern variations in control object (i.e., motions into, out of or relative to virtual construct 198B) as indicating control or other useful information. A gesture trainer 198C and gesture properties extractor 198D provide functionality to define, build and/or customize gesture properties 198A.

[0132] A context determiner 198G and object of interest determiner 198H provide functionality to determine from the object, position, motion and attribute information objects of interest (e.g., control objects, or other objects to be modeled and analyzed), objects not of interest (e.g., background) based upon a detected context. For example, when the context is determined to be an identification context, a human face will be determined to be an object of interest to the system and will be determined to be a control object. On the other hand, when the context is determined to be a fingertip control context, the finger tips will be determined to be object(s) of interest and will be determined to be a control objects whereas the user's face will be determined not to be an object of interest (i.e., background). Further, when the context is determined to be a styli (or other tool) held in the fingers of the user, the tool tip will be determined to be object of interest and a control object whereas the user's fingertips might be determined not to be objects of interest (i.e., background). Background objects can be included in the environmental information provided to environmental filter 197H of model management module 197.

[0133] A virtual environment manager 198E provides creation, selection, modification and de-selection of one or more virtual constructs 198B (see FIGS. 8-3, 8-4). In some implementations, virtual constructs (e.g., a virtual object defined in space; such that variations in real objects relative to the virtual construct, when detected, can be interpreted for control or other purposes (see FIGS. 8-3, 8-4)) are used to determine variations (i.e., virtual “contact” with the virtual construct, breaking of virtual contact, motion relative to a construct portion, etc.) to be interpreted as engagements, dis-engagements, motions relative to the construct(s), and so forth, enabling the system to interpret pinches, pokes and grabs, and so forth. Interaction interpretation module 198 provides as output the command information, related information and other information discernable from the object, position, motion and attribute information that might be useful in controlling a machine from recognition engine 198F to an application control system 90D.

[0134] Further with reference to FIG. 20, an application control system 90D includes a control module 199 that provides functionality to determine and authorize commands based upon the command and other information obtained from interaction system 90C.

[0135] A control module 199 implementation comprises a command engine 199F to determine whether to issue command(s) and what command(s) to issue based upon the command information, related information and other information discernable from the object, position, motion and attribute information, as received from an interaction interpretation module 198. Command engine 199F employs command/control repository 199A (e.g., application commands, OS commands, commands to MSCS, misc. commands) and related information indicating context received from the interaction interpretation module 198 to determine one or more commands corresponding to the gestures, context, etc. indicated by the command information. For example, engagement gestures can be mapped to one or more controls, or a control-less screen location, of a presentation device associated with a machine under control. Controls can include imbedded controls (e.g., sliders, buttons, and other control objects in an application), or environmental level controls (e.g., windowing controls, scrolls within a window, and other controls affecting the control environment). In implementations, controls may be displayed using 2D presentations (e.g., a cursor, cross-hairs, icon, graphical representation of the control object, or other displayable object) on display screens and/or presented in 3D forms using holography, projectors or other mechanisms for creating 3D presentations, or audible (e.g., mapped to sounds, or other mechanisms for conveying audible information) and/or touchable via haptic techniques.

[0136] Further, an authorization engine 199G employs biometric profiles 199B (e.g., users, identification information, privileges, etc.) and biometric information received from the interaction interpretation module 198 to determine whether commands and/or controls determined by the command engine 199F are authorized. A command builder 199C and biometric profile builder 199D provide functionality to define, build and/or customize command/control repository 199A and biometric profiles 199B.

[0137] Selected authorized commands are provided to machine(s) under control (i.e., “client”) via interface layer 196. Commands/controls to the virtual environment (i.e., interaction control) are provided to virtual environment manager 198E. Commands/controls to the emission/detection systems (i.e., sensory control) are provided to emission module 91 and/or detection module 92 as appropriate.

[0138] In various implementations and with reference to FIG. 23A, 23B, a Machine Sensory Controller System 189 can be embodied as a standalone unit(s) 189-1 coupleable via an interface (e.g., wired or wireless)), embedded (e.g., within a machine 188-1, 188-2 or machinery under control) (e.g., FIG. 23A: 189-2, 189-3, FIG. 23B: 189B) or combinations thereof.

[0139] FIG. 24 illustrates an example computing system that can comprise one or more of the elements shown in FIGS. 16A and 16B. In particular, FIG. 24 illustrates an exemplary computing system 2400, such as a PC (or other suitable “processing” system), that can comprise one or more of the MSCS elements shown in FIGS. 17-20 according to an implementation. While other application-specific device/process alternatives might be utilized, such as those already noted, it will be presumed for clarity sake that systems 90A-90D elements (FIGS. 17-20) are implemented by one or more processing systems consistent therewith, unless otherwise indicated.

[0140] As shown, computer system 2400 comprises elements coupled via communication channels (e.g. bus 2401) including one or more general or special purpose processors 2402, such as a Pentium® or Power PC®, digital signal processor (“DSP”), or other processing. System 2400 elements also include one or more input devices 2403 (such as a mouse, keyboard, joystick, microphone, remote control unit, tactile, biometric or other sensors 240x, and so on), and one or more output devices 2404, such as a suitable display, joystick feedback components, speakers, biometric or other actuators, and so on, in accordance with a particular application.

[0141] System 2400 elements also include a computer readable storage media reader 2405 coupled to a computer readable storage medium 2406, such as a storage/memory device or hard or removable storage/memory media; examples are further indicated separately as storage device 2408 and non-transitory memory 2409, which can include hard disk variants, floppy/compact disk variants, digital versatile disk (“DVD”) variants, smart cards, read only memory, random access memory, cache memory or others, in accordance with a particular application (e.g. see data store(s) 197A, 198A, 199A and 199B of FIG. 20). One or more suitable communication devices 2407 can also be included, such as a modem, DSL, infrared, etc. for providing inter-device communication directly or via suitable private or public networks, such as the Internet. Working memory 2409 is further indicated as including an operating system (“OS”) 2491, predictive discrepancy determiner 2413 and other programs 2492, such as application programs, mobile code, data, or other information for implementing systems 90A-90D elements, which might be stored or loaded therein during use.

[0142] System 2400 element implementations can include hardware, software, firmware or a suitable combination. When implemented in software (e.g. as an application program, object, downloadable, servlet, and so on, in whole or part), a system 900 element can be communicated transitionally or more persistently from local or remote storage to memory for execution, or another suitable mechanism can be utilized, and elements can be implemented in compiled, simulated, interpretive or other suitable forms. Input, intermediate or resulting data or functional elements can further reside more transitionally or more persistently in a storage media or memory, (e.g. storage device 2408 or memory 2409) in accordance with a particular application.

[0143] Certain potential interaction determination, virtual object selection, authorization issuances and other aspects enabled by input/output processors and other element implementations disclosed herein can also be provided in a manner that enables a high degree of broad or even global applicability; these can also be suitably implemented at a lower hardware/software layer. Note, however, that aspects of such elements can also be more closely linked to a particular application type or machine, or might benefit from the use of mobile code, among other considerations; a more distributed or loosely coupled correspondence of such elements with OS processes might thus be more desirable in such cases.

[0144] Referring to FIG. 25, which illustrates a system for capturing image data according to one implementation of the technology disclosed. System 2500 is preferably coupled to a wearable device 2501 that can be a personal head mounted display (HMD) having a goggle form factor such as shown in FIG. 25, a helmet form factor, or can be incorporated into or coupled with a watch, smartphone, or other type of portable device.

[0145] In various implementations, the system and method for capturing 3D motion of an object as described herein can be integrated with other applications, such as a head-mounted device or a mobile device. Referring again to FIG. 25, a head-mounted device 2501 can include an optical assembly that displays a surrounding environment or a virtual environment to the user; incorporation of the motion-capture system 2500 in the head-mounted device 2501 allows the user to interactively control the displayed environment. For example, a virtual environment can include virtual objects that can be manipulated by the user's hand gestures, which are tracked by the motion-capture system 2500. In one implementation, the motion-capture system 2500 integrated with the head-mounted device 2501 detects a position and shape of user's hand and projects it on the display of the head-mounted device 2500 such that the user can see her gestures and interactively control the objects in the virtual environment. This can be applied in, for example, gaming or internet browsing.

[0146] In one embodiment, information about the interaction with a virtual object can be shared by a first HMD user with a HMD of a second user. For instance, a team of surgeons can collaborate by sharing with each other virtual incisions to be performed on a patient. In some embodiments, this is achieved by sending to the second user the information about the virtual object, including primitive(s) indicating at least one of a type, size, and/or features and other information about the calculation point(s) used to detect the interaction. In other embodiments, this is achieved by sending to the second user information about the predictive model used to track the interaction.

[0147] System 2500 includes any number of cameras 2502, 2504 coupled to sensory processing system 2506. Cameras 2502, 2504 can be any type of camera, including cameras sensitive across the visible spectrum or with enhanced sensitivity to a confined wavelength band (e.g., the infrared (IR) or ultraviolet bands); more generally, the term “camera” herein refers to any device (or combination of devices) capable of capturing an image of an object and representing that image in the form of digital data. For example, line sensors or line cameras rather than conventional devices that capture a two-dimensional (2D) image can be employed. The term “light” is used generally to connote any electromagnetic radiation, which may or may not be within the visible spectrum, and may be broadband (e.g., white light) or narrowband (e.g., a single wavelength or narrow band of wavelengths).

[0148] Cameras 2502, 2504 are preferably capable of capturing video images (i.e., successive image frames at a constant rate of at least 15 frames per second); although no particular frame rate is required. The capabilities of cameras 2502, 2504 are not critical to the technology disclosed, and the cameras can vary as to frame rate, image resolution (e.g., pixels per image), color or intensity resolution (e.g., number of bits of intensity data per pixel), focal length of lenses, depth of field, etc. In general, for a particular application, any cameras capable of focusing on objects within a spatial volume of interest can be used. For instance, to capture motion of the hand of an otherwise stationary person, the volume of interest might be defined as a cube approximately one meter on a side.

[0149] As shown, cameras 2502, 2504 can be oriented toward portions of a region of interest 2512 by motion of the device 2501, in order to view a virtually rendered or virtually augmented view of the region of interest 2512 that can include a variety of virtual objects 2516 as well as contain an object of interest 2514 (in this example, one or more hands) moves within the region of interest 2512. One or more sensors 2508, 2510 capture motions of the device 2501. In some implementations, one or more light sources 2515, 2517 are arranged to illuminate the region of interest 2512. In some implementations, one or more of the cameras 2502, 2504 are disposed opposite the motion to be detected, e.g., where the hand 2514 is expected to move. This is an optimal location because the amount of information recorded about the hand is proportional to the number of pixels it occupies in the camera images, and the hand will occupy more pixels when the camera's angle with respect to the hand's “pointing direction” is as close to perpendicular as possible. Sensory processing system 2506, which can be, e.g., a computer system, can control the operation of cameras 2502, 2504 to capture images of the region of interest 2512 and sensors 2508, 2510 to capture motions of the device 2501. Information from sensors 2508, 2510 can be applied to models of images taken by cameras 2502, 2504 to cancel out the effects of motions of the device 2501, providing greater accuracy to the virtual experience rendered by device 2501. Based on the captured images and motions of the device 2501, sensory processing system 2506 determines the position and/or motion of object 2514.

[0150] For example, as an action in determining the motion of object 2514, sensory processing system 2506 can determine which pixels of various images captured by cameras 2502, 2504 contain portions of object 2514. In some implementations, any pixel in an image can be classified as an “object” pixel or a “background” pixel depending on whether that pixel contains a portion of object 2514 or not. Object pixels can thus be readily distinguished from background pixels based on brightness. Further, edges of the object can also be readily detected based on differences in brightness between adjacent pixels, allowing the position of the object within each image to be determined. In some implementations, the silhouettes of an object are extracted from one or more images of the object that reveal information about the object as seen from different vantage points. While silhouettes can be obtained using a number of different techniques, in some implementations, the silhouettes are obtained by using cameras to capture images of the object and analyzing the images to detect object edges. Correlating object positions between images from cameras 2502, 2504 and cancelling out captured motions of the device 2501 from sensors 2508, 2510 allows sensory processing system 2506 to determine the location in 3D space of object 2514, and analyzing sequences of images allows sensory processing system 2506 to reconstruct 3D motion of object 2514 using conventional motion algorithms or other techniques. See, e.g., U.S. patent application Ser. No. 13/414,485 (filed on Mar. 7, 2012) and U.S. Provisional Patent Application Nos. 61/724,091 (filed on Nov. 8, 2012) and 61/587,554 (filed on Jan. 7, 2012), the entire disclosures of which are hereby incorporated by reference.

[0151] Presentation interface 2520 employs projection techniques in conjunction with the sensory based tracking in order to present virtual (or virtualized real) objects (visual, audio, haptic, and so forth) created by applications loadable to, or in cooperative implementation with, the device 2501 to provide a user of the device with a personal virtual experience. Projection can include an image or other visual representation of an object.

[0152] One implementation uses motion sensors and/or other types of sensors coupled to a motion-capture system to monitor motions within a real environment. A virtual object integrated into an augmented rendering of a real environment can be projected to a user of a portable device 2501. Motion information of a user body portion can be determined based at least in part upon sensory information received from imaging 2502, 2504 or acoustic or other sensory devices. Control information is communicated to a system based in part on a combination of the motion of the portable device 2501 and the detected motion of the user determined from the sensory information received from imaging 2502, 2504 or acoustic or other sensory devices. The virtual device experience can be augmented in some implementations by the addition of haptic, audio and/or other sensory information projectors. For example, an optional video projector 2520 can project an image of a page (e.g., virtual device) from a virtual book object superimposed upon a real world object, e.g., desk 2516 being displayed to a user via live video feed; thereby creating a virtual device experience of reading an actual book, or an electronic book on a physical e-reader, even though no book nor e-reader is present. Optional haptic projector can project the feeling of the texture of the “virtual paper” of the book to the reader's finger. Optional audio projector can project the sound of a page turning in response to detecting the reader making a swipe to turn the page. Because it is a virtual reality world, the back side of hand 2514 is projected to the user, so that the scene looks to the user as if the user is looking at the user's own hand(s).

[0153] A plurality of sensors 2508, 2510 coupled to the sensory processing system 2506 to capture motions of the device 2501. Sensors 2508, 2510 can be any type of sensor useful for obtaining signals from various parameters of motion (acceleration, velocity, angular acceleration, angular velocity, position/locations); more generally, the term “motion detector” herein refers to any device (or combination of devices) capable of converting mechanical motion into an electrical signal. Such devices can include, alone or in various combinations, accelerometers, gyroscopes, and magnetometers, and are designed to sense motions through changes in orientation, magnetism or gravity. Many types of motion sensors exist and implementation alternatives vary widely.

[0154] The illustrated system 2500 can include any of various other sensors not shown in FIG. 25 for clarity, alone or in various combinations, to enhance the virtual experience provided to the user of device 2501. For example, in low-light situations where free-form gestures cannot be recognized optically with a sufficient degree of reliability, system 2506 may switch to a touch mode in which touch gestures are recognized based on acoustic or vibrational sensors. Alternatively, system 2506 may switch to the touch mode, or supplement image capture and processing with touch sensing, when signals from acoustic or vibrational sensors are sensed. In still another operational mode, a tap or touch gesture may act as a “wake up” signal to bring the image and audio analysis system 2506 from a standby mode to an operational mode. For example, the system 2506 may enter the standby mode if optical signals from the cameras 2502, 2504 are absent for longer than a threshold interval.

[0155] It will be appreciated that the figures shown in FIG. 25 are illustrative. In some implementations, it may be desirable to house the system 2500 in a differently shaped enclosure or integrated within a larger component or assembly. Furthermore, the number and type of image sensors, motion detectors, illumination sources, and so forth are shown schematically for the clarity, but neither the size nor the number is the same in all implementations.

[0156] FIG. 29 shows a flowchart 2900 of initializing orientation of a three-dimensional (3D) model of an object. Flowchart shown in FIG. 29 can be implemented at least partially with by one or more processors configured to receive or retrieve information, process the information, store results, and transmit the results. Other implementations may perform the actions in different orders and/or with different, varying, alternative, modified, fewer or additional actions than those illustrated in FIG. 29. Multiple actions can be combined in some implementations. For convenience, this flowchart is described with reference to the system that carries out a method. The system is not necessarily part of the method.

[0157] At action 2902, at least one three-dimensional (3D) model of an object is accessed along with observed information of the object movable in space. In one implementation, a primary orientation parameter of the model from the observed information is determined, as described below.

[0158] At action 2912, contours of the object in the observed information are detected and a representative normal to the detected contours is calculated.

[0159] The method further includes filtering the contours to detect edges and calculating the representative normal from the filtered contours. It also includes using brightness gradients to calculate the representative normal. It includes using brightness contours to calculate the representative normal. It includes averaging a plurality of normals to calculate the representative normal. In another implementation, it includes calculating an argument value for a histogram of a plurality of normals and using the argument value as the representative normal.

[0160] In some other implementations, the method further includes identifying a plurality of normals to the contours of the object and determining a normal from the plurality with a maximum magnitude and selecting the normal with the maximum magnitude as the representative normal. In one implementation, it includes identifying a plurality of normals to the contours of the object and selecting, as the representative normal, a normal from the plurality with a magnitude determined from magnitudes of the plurality and a direction determined from the plurality. The determining further includes determining an average of the magnitudes and the directions of the normals. Also, the determining further includes determining a mean of the magnitudes and the directions of the normals.

[0161] At action 2922, a vector representing a 3D angle from the object to a point of observation is accessed. One example of such a point of observation is a camera or an illumination source.

[0162] At action 2932, a primary orientation of the object is determined as a cross-product of the representative normal and the vector.

[0163] At action 2942, the calculated primary orientation parameter is used to initialize the model.

[0164] In yet other implementations, the method includes initializing a model portion by applying the primary orientation parameter to at least a portion of the model and aligning the portion to an observed orientation of the object. In one implementation, the object is a human body. In another implementation, the object is a hand. In a further implementation, the object is a tool.

[0165] In some implementations, the method further includes modifying power consumption profiles of a machine sensory control system responsive to initialization of the model.

[0166] This method and other implementations of the technology disclosed can include one or more of the following features and/or features described in connection with additional methods disclosed. In the interest of conciseness, the combinations of features disclosed in this application are not individually enumerated and are not repeated with each base set of features. The reader will understand how features identified in this section can readily be combined with sets of base features identified as implementations in sections of this application.

[0167] Other implementations can include a non-transitory computer readable storage medium storing instructions executable by a processor to perform any of the methods described above. Yet another implementation can include a system including memory and one or more processors operable to execute instructions, stored in the memory, to perform any of the methods described above.

[0168] FIG. 30 shows a flowchart 3000 of initializing orientation of a three-dimensional (3D) model of an object. Flowchart shown in FIG. 30 can be implemented at least partially with by one or more processors configured to receive or retrieve information, process the information, store results, and transmit the results. Other implementations may perform the actions in different orders and/or with different, varying, alternative, modified, fewer or additional actions than those illustrated in FIG. 30. Multiple actions can be combined in some implementations. For convenience, this flowchart is described with reference to the system that carries out a method. The system is not necessarily part of the method.

[0169] At action 3002, at least one three-dimensional (3D) model of an object is accessed along with observed information of the object movable in space. In one implementation, a plurality of initialization parameters of the model from the observed information is determined, as described below.

[0170] At action 3012, observed extremities of the object and corresponding parts of the model are determined.

[0171] At action 3022, a determination is made whether the model has more extremities than that of the observed information.

[0172] At action 3032, at least one extremity correction is applied to the model that bends or folds some of the extremities in the model out of view.

[0173] At action 3042, an extremity correction is selected that fits the observed information about the extremities of the object.

[0174] In some implementations, applying at least one extremity correction further includes curving one or more model extremities under a model portion to model an occlusion.

[0175] This method and other implementations of the technology disclosed can include one or more of the following features and/or features described in connection with additional methods disclosed. In the interest of conciseness, the combinations of features disclosed in this application are not individually enumerated and are not repeated with each base set of features. The reader will understand how features identified in this section can readily be combined with sets of base features identified as implementations in sections of this application.

[0176] Other implementations can include a non-transitory computer readable storage medium storing instructions executable by a processor to perform any of the methods described above. Yet another implementation can include a system including memory and one or more processors operable to execute instructions, stored in the memory, to perform any of the methods described above.

[0177] FIG. 31 shows a flowchart 3100 of initializing orientation of a three-dimensional (3D) model of an object using a. Flowchart shown in FIG. 31 can be implemented at least partially with by one or more processors configured to receive or retrieve information, process the information, store results, and transmit the results. Other implementations may perform the actions in different orders and/or with different, varying, alternative, modified, fewer or additional actions than those illustrated in FIG. 31. Multiple actions can be combined in some implementations. For convenience, this flowchart is described with reference to the system that carries out a method. The system is not necessarily part of the method.

[0178] At action 3102, at least one three-dimensional (3D) model of an object is accessed along with observed information of the object movable in space. In one implementation, a primary orientation parameter of the model from the observed information is determined, as described below.

[0179] At action 3112, a cloud of points in the observed information is detected and a plane that best fits the cloud of points is calculated using a statistical best fit method.

[0180] At action 3122, an orientation vector is calculated that is normal to the plane that best fits the cloud of points.

[0181] At action 3132, the calculated orientation vector is used as a parameter to initialize the model.

[0182] At action 3142, contours of the object in the observed information and calculating a representative normal to the detected contours are detected, a vector representing a 3D angle from the object to a point of observation is accessed, a secondary orientation vector of the object is calculated as a cross-product of the representative normal and the vector, and the calculated secondary orientation vector is used as an additional parameter to initialize the model.

[0183] This method and other implementations of the technology disclosed can include one or more of the following features and/or features described in connection with additional methods disclosed. In the interest of conciseness, the combinations of features disclosed in this application are not individually enumerated and are not repeated with each base set of features. The reader will understand how features identified in this section can readily be combined with sets of base features identified as implementations in sections of this application.

[0184] Other implementations can include a non-transitory computer readable storage medium storing instructions executable by a processor to perform any of the methods described above. Yet another implementation can include a system including memory and one or more processors operable to execute instructions, stored in the memory, to perform any of the methods described above.

[0185] Reference in the specification to “one implementation” or “an implementation” means that a particular feature, structure, or characteristic described in connection with the implementation is included in at least one implementation of the technology disclosed. The appearances of the phrase “in one implementation” in various places in the specification are not necessarily all referring to the same implementation.

[0186] While the technology disclosed has been described by way of example and in terms of the specific implementations, it is to be understood that the technology disclosed is not limited to the disclosed implementations. To the contrary, it is intended to cover various modifications and similar arrangements as would be apparent to those skilled in the art. Therefore, the scope of the appended claims should be accorded the broadest interpretation so as to encompass all such modifications and similar arrangements.