US 5,909,504 AGrant
Method of Testing a Machine Vision Inspection System
Issue Date:1999-06-01
•20 Claims
•6 Drawing Sheets
Abstract
The invention provides a method of testing a machine vision system of the type that inspects a feature (e.g., an electronic component) using object-oriented constructs that instantiate an inspection object from an inspection class that is associated with a type of the feature (e.g., the rectilinear component) and that invoke a method member of that object to inspect the feature to determine its characteristics (e.g., position, angular orientation, and shape conformance). The method of the invention includes the steps of instantiating a test object from a test class that corresponds to the inspection class, invoking a method member of that test object to generate one or more test images representing the feature, inspecting the test images to determine characteristics of the features therein, and reporting results of those inspections.
Metadata
Assignee
- Cognex Corporation
Inventor
- Steven M. Whitman
Application Information
Application Number:US 618460&
Filing Date:1996-03-15
Priority Date:1996-03-15
Art Unit:271
Classifications
IPC:
G06F 1100G06T 760
Field of Search:
382348356364437250395707371702141;144;145;146;147;148;149;150;151;152;100;159;160;161;30987;126;130;92;86237468.15;468.16;468.17;468.28;481;488;551.01;552;578;571.018559.46;559.39183.01;183.09;183.13;183.14;70410327.4;71;72;22.1;27.1;48;67.135;82;155;157;158;159;166
Patent Drawings (6 sheets)
Description
RESERVATION OF COPYRIGHT
The disclosure of this patent document contains material which is subject to copyright protection. The owner thereof has no objection to facsimile reproduction by anyone of the patent document or of the patent disclosure, as it appears in the United States Patent and Trademark Office patent file or records, but otherwise reserves all rights under copyright law.
BACKGROUND OF THE INVENTION
The invention pertains to machine vision and, more particularly, to methods for testing machine vision inspection systems.
In automated assembly processes, such as those for assembling electronic circuit boards, it is important to determine the exact location of components prior to their placement for final assembly. For example, an assembly system must know the position and angular orientation of an electronic component before placing and soldering its leads to a printed circuit board.
Accuracy in positioning is ever more critical as the size of electronic components decrease and the number of conductive leads increase. Such is the case with packaged integrated circuits, which may have hundreds of fine wire leads that must be aligned with pads on a printed circuit board.
Component inspection is important in other automated assembly processes, as well. For example, in an automotive assembly line, welds and rivets must be verified for safety. Likewise, in a pharmaceutical assembly line, the placement of caps and seals on bottles must be confirmed to minimize the risks of spoilage and tampering.
The art has developed a variety of systems for facilitating automated assembly processes. The most promising of these are "machine vision" systems that analyze video images of parts under assembly to verify their integrity and placement. A machine vision system for surface mounted device inspection, for example, analyzes images from video cameras on robotic assembly equipment to determine and coordinate the precise placement of electronic circuit components.
Notwithstanding the precision of the conveyor belts, robotic arms and other automated assembly equipment, components may wind somewhat out of position or slightly misshapen as they are being readied for assembly. A packaged integrated circuit that is being aligned for soldering to a printed circuit board, for example, may be slightly skewed and have one short lead. A machine vision system must be able to recognize the components and be able to determine their positions in spite of such deviations. Where the deviations are too severe, however, these systems must signal an alert so that the process can be aborted.
Traditional machine vision systems for surface mounted device inspection rely on standardized libraries to facilitate component identification. More advanced systems, such as those sold by the assignee hereof (Cognex Corporation) under the trade names SMD2 and SMD PGP, permit users to define components to be inspected, using a description language (UDL) of the type described in U.S. Pat. No. 5,371,690.
Both conventional and advanced surface mounted device systems permit users to define "characteristics," such as deviations in position and angular orientation, that are acceptable for the assembly process. For example, the systems can be set up to identify and locate a component (e.g., a large-leaded rectangular component) and to signal an alert only if its position deviates by more than 0.5 centimeters from center or its angular orientation deviates by more than 5 degrees from horizontal. The more advanced systems permit users to define additional characteristics, such as variations in lead angles and lengths, body colors or intensities, and edge polarities, that are acceptable for the assembly process.
With the advent of object-oriented programming, many of the machine vision inspection systems use "class" constructs to segregate component definitions and permissible characteristics by component type. The aforementioned SMD PGP product, for example, relies on separate classes to segregate and store data and method members necessary for inspection ball grid arrays, rectilinear devices, etc.
Before machine vision inspections systems can be put to use for surface mounted device or other types of inspections, the systems must be tested to ensure that the definitions of permissible characteristics are consistent with those mandated by the assembly equipment and quality control standards. For example, if a pick-up nozzle cannot adequately grip a component that is rotated more than 7 degrees, the user must be apprised of this so that he or she can modify the definitions accordingly.
According to the prior art, testing is usually accomplished by arranging "by hand" a test component in various positions and running the inspection to determine whether it returns proper results. An alternative is to compile during runtime a database of images and to run the inspection system on each one of those images. A drawback of these prior art procedures is the difficulty in testing all possible variations. Another drawback is the time involved in creating the image database and in performing tests "by hand."
In view of the foregoing, an object of this invention is to provide improved methods for machine vision analysis and, particularly, improved methods for testing machine vision inspection systems.
More particularly, an object of the invention is to provide methods for testing machine vision inspection systems to evaluate their operation under a wide range of operational circumstances.
Yet another object of the invention is to provide methods for testing machine vision surface mounted device inspection systems.
A related object is to provide methods for generating images that can be used in testing machine vision inspection systems.
Yet still another object of the invention is to provide such methods that can execute quickly, and without undue consumption of resources, on a wide range of machine vision analysis equipment.
SUMMARY OF THE INVENTION
The foregoing objects are attained by the invention which provides, in one aspect, a method of testing a machine vision system of the type that inspects an image feature (e.g., an electronic component) using object-oriented constructs that (1) instantiate an inspection object from an inspection class associated with a type of the feature (e.g., the rectilinear component) and (2) invoke a method member of that object to inspect the feature to determine its characteristics (e.g., position, angular orientation, and shape conformance). The method is characterized by the steps of instantiating a test object from a test class that inherits from the inspection class and from a test harness class, invoking a method member of that test object to generate one or more test images representing the feature, using the inspect method of that object to inspect the test images to determine characteristics of the features therein, and reporting results of those inspections.
A method according to this aspect of the invention can be used, for example, to test a machine vision system that inspects fiducial marks. In this regard, the method instantiates a fiducial test object from a test class that inherits from the inspection class used by the machine vision system for inspection of such marks. A method member in that test object generates one or more test images representing the fiducial mark, e.g., in various positions, orientations and sizes. Another method member in that test object (i.e., a method member inherited from the inspection class) is called to inspect those test images to determine those positions and orientations, as well as to determine the acceptability (or quality) of the marks. Results of those inspections are then reported, e.g., to the user, for evaluation of the fiducial inspection class and, more generally, of the machine vision inspection system.
According to further aspects, the invention provides methods as described above in which the test images depict the feature (e.g., circuit component) with a characteristic whose "value" is selected from multiple potential values. For example, a test image can depict the feature at any one of many possible positions expected during operation of the assembly process. The feature may also be depicted with other characteristics, such as variations in angular orientation, size, and grey value intensity (color).
In related aspects, the invention provides methods as described above in which the test object generates multiple test images, each representing the feature with multiple characteristics that have values selected from a range of potential values. For example, test images can be generated to represent the feature in the full range of positions and angular orientations that may be expected during the assembly process. The values for each of the characteristics can be selected in a predetermined sequence or using fixed values, though they are preferably selected at random.
According to further aspects of the invention, the method provides for instantiating the test object from a test class that inherits method and data members from the inspection class, as well as from a test harness class. Such "multiple inheritance" affords the test object access to the feature definition provided in the inspection class. The test object can make use of this by generating the test images generated with characteristics within the full range of expected values permitted by that definition.
In a related aspect, the invention provides a method as described above in which the inspection results are reported using members inherited by the test object from the test harness class.
Further aspects of the invention provide methods as described above in which the test image inspection results are compared with expected results. The results of those comparisons can, moreover, be tracked to facilitate evaluation of the inspection model.
Still further aspects of the invention provide methods for generating test images in accord with the techniques described above.
Yet still further aspects of the invention provide methods for testing machine vision systems for surface mounted device inspection utilizing the foregoing methods.
These and other aspects of the invention are evident in the drawings and the descriptions and claims that follow.
As those skilled in the art will appreciate from the discussion herein, the invention has wide application in industrial and research applications. It can be used to test and evaluate rapidly machine vision inspection systems of the type, e.g., used in automated circuit assembly lines. Results of such testing can be used to improve the inspection systems, and, thereby the speed and quality of the assembly processes.
BRIEF DESCRIPTION OF THE DRAWINGS
A more complete understanding of the invention may be attained by reference to the drawings, in which
FIG. 1 depicts a machine vision system for use in practice of the invention;
FIG. 2 depicts inheritance relationships between inspection classes of a machine vision inspection system and test classes used in a method according to the invention; and
FIG. 3 depicts a methodology according to the invention for testing a machine vision inspection system.
FIGS. 4-6 depict a class hierarchy for use in a method according to the invention, as well as a relationship between that hierarchy and the hierarchy of the inspection classes.
Index of Terminology
To facilitate understanding of the invention, and without limitation, the following definitions are provided:
"inspection class" means an object-oriented programming class to be tested containing data members and/or method members associated with an image feature;
"inspection method" means a method member of an inspection class;
"method member" means a method (or method function) contained within a class or object;
"multiply inherit" means to derive characteristics and/or behaviors (i.e., data members and/or method members) from multiple base classes;
"test class" means a class containing data members and/or method members inherited at least from an inspection class;
"test harness" means an automated methodology for testing software (e.g., an inspection class);
"test image" means an image (e.g., generated by a test object) representing an image feature of the type inspected by the inspection class being tested; and
"test object" means an instance (or instantiation) of a test class.
DETAILED DESCRIPTION OF THE ILLUSTRATED EMBODIMENT
FIG. 1 illustrates a system 5 for machine vision inspection and for testing thereof. The system 5 for machine vision inspection includes a capturing device 10 that generates an image of a scene including an object 1. The image comprises pixel data representing, in the conventional manner, the intensities or colors of the scene. The image data is transmitted from capturing device via communication path 11 to image analyses system 12 including central processing unit 13, main memory 14, input/output system 15 and disk drive 16, all of the conventional type. The system 12 is a conventional digital data processor, or a vision processing system of the type commercially available from the assignee hereof (Cognex Corporation) programmed for machine vision inspection in accord with prior art techniques.
The system 12 and, more particularly, central processing unit 13, is additionally programmed in accord with the teachings hereof for testing the components (e.g., inspection classes) of the machine vision inspection system that analyze images captured by device 10.
FIG. 2 depicts a relationship between inspection classes of the machine vision inspection system and those of the present invention for testing that system. The inspection classes 30, 32, 34 define the features (e.g., circuit components) to be inspected, e.g., by way of standardized libraries or by way of a UDL of the type disclosed in aforementioned U.S. Pat. No. 5,371,690, the teachings of which are incorporated herein by reference. The inspection classes 30, 32, 34 also define characteristics of those features, such as their permissible positions, angular orientations, sizes, and colors or intensities. In instances where the feature has defined sub-features, such as component lead lengths and positions, the inspection classes define the characteristics of those sub-features.
In accord with current programming practices, the feature definitions and corresponding characteristics are "cloaked" in object-oriented constructs of the type available in commercially available software language compilers for C++, Turbo Pascal, etc. Although machine vision inspection systems utilizing object-oriented constructs are understood to be commercially available from a number of suppliers, a preferred such system is the surface mounted device inspection system, referred to as SMD PGP, available from the assignee hereof.
Each of the illustrated inspection classes pertains to a specific type of feature (or component) to be inspected. For example, class 30 contains definitions and characteristics for inspection of fiducial images; class 32, for ball grid array devices; and, class 34, for rectilinear circuit devices.
The illustrated test system includes four test classes 36, 38, 40, 42. Together, these define a test harness that tracks and reports results of testing the machine vision inspection system. As shown in the illustration, class 42 inherits from class 40, which in turn inherits from class 38, which in turn inherits from class 36.
Class 36 provides the base functionality for testing. Its primary use is as an interface specification that permits the test system to be treated in a uniform manner. Class 38 adds data and method members for images, translation and rotation limits, centers, statistics about inspection performance, and additional reporting. Class 40 adds data and method members for specific support of device inspection, while class 42 adds data and method members for testing inspection of a specific type of feature (or component), to wit, rectilinear devices.
As shown in the illustration, test class 44 inherits data and method members from corresponding inspection and test harness classes. Particularly, test class 44 inherits from inspection class 34 and test harness class 42, both of which relate to rectilinear devices. As discussed below, the test class 44 is used to instantiate test objects that generate images of such devices for evaluation by an inspection object that is, effectively, instantiated from the test class 44 (via inheritance from the inspection class 34). Although not illustrated, test harness class 42 can have peers that define data and method members for testing other features, such as fiducial marks and ball grid arrays. Further test classes (not shown) inherit from those peers and from the corresponding inspection classes 30, 32, in like manner to that described above. Also not illustrated, test harness classes 36 and 38 can have peers that define data and method members for testing other features, such as calibrations and circuit board inspections.
FIG. 3 depicts a preferred methodology accordingly to the invention for testing a machine vision inspection system. In step 50, the method instantiates a test object from test class 44, which inherits from rectilinear feature inspection class 34. It will be appreciated that inspection classes for a full range of circuit components and other features can be tested in accord with this same illustrated methodology.
In step 52, the method determines a position and angular orientation of a "test" feature to be incorporated into a test image. Potential values for these characteristics are determined from the definitions supplied by the inspection class 34. For example, where that inspection class defines the feature as residing within one centimeter of a reference point, the step 52 selects an "acceptable" position from within that range or an "unacceptable" position outside it. The selection can be at random, in sequence (in instances where multiple test images are generated) or using fixed values. In addition to position and angular orientation, it will be appreciated that other characteristics of the test feature can be generated in step 52 as well.
In step 54, the method initializes a test image. This includes, for example, generating an artificial background simulating the image background expected during the actual assembly process. This also includes generating a base test feature having the characteristics determined in step 52.
In steps 56-60, the method adds additional characteristics to the test feature in order to better simulate an image of the type to be inspected during actual assembly. Particularly, in step 56, the method selects a sub-feature, such as a conductive lead array, to be added to the test feature. In step 58, the method determines a value of a characteristic of that sub-feature. For example, where the sub-feature is a lead array, the method can select a length of that array. Likewise, it can select an angle for specific leads therein. As above, the selection is made in accord with definitions supplied by the inspection class 34. A random selection can be made from among the permissible (and impermissible) values, or the selection can be made in succession (in instances where multiple images are generated) or the selection can be made using fixed values. In step 60, the method adds the sub-feature, with its characteristic, to the test image.
Once the test image is completed, step 64 uses the inspection method (inherited from the inspection class 34) to inspect the test image. The inspection is carried out in the normal course and returns, for example, the position and angular orientation of the feature in the test image, as well as an indication of whether the inspection was a success (e.g., whether the feature was "complete" as defined by the inspection class).
In step 66, the method compares the results of the inspection with the expected results. For example, the method compares the known position of the test feature in the test image with that returned by the inspection. Likewise, it compares the known angular orientation of the test feature with that returned by the inspection. Furthermore, it compares the expected success of the inspection with that returned by the inspection. In accord with methods defined by the test harness classes, the results of these comparisons can be tracked and reported to the user for evaluation of the inspection class 34 and, more generally, the machine vision inspection system; see, step 68.
Preferably, steps 52-68 are run repeatedly in order to generate multiple synthetic images, each with randomly varying characteristics. After a suitable number of runs (as determined in accord with the complexity of the feature being inspected) any deficiencies in the inspection class will become readily apparent.
A further understanding of the illustrated method for testing a machine vision inspection system may be attained by reference to the description that follows. This pertains to a preferred embodiment for use in connection with the aforementioned SMD PGP product. In that description, the preferred embodiment is referred to by the term "Monte Carlo."
Monte Carlo Inspection Overview
Monte Carlo testing is a technique that uses randomly generated data as inputs to a process under test. Since the testing framework knows what random data was generated, it can test the outputs of the software to verify that the correct actions occurred.
FIG. 4 is a partial class hierarchy of exemplary Monte Carlo classes. This diagram is a representation of the types of class hierarchies that can be developed. In practice, the actual class hierarchies can differ (except for the bases class ccMCTestable).
A user of Monte Carlo would use one of the above classes (or create a new one) to multiply inherit with an inspection (or other) terminal class. For example, if the developer of a small leaded inspection class (ccSmallLeadedInspection) inherited from the ccRectDeviceInspection class then they would also inherit from the ccMCRectilinearInspection class if they wished to perform Monte Carlo testing.
The terminal classes in the hierarchy contain most of the necessary methods to perform complete Monte Carlo testing. For example, the ccMCRectilinearInspection contains the methods to generate randomized placements of a synthetically generated image, run the inspection and gather the statistics. This is typical of the other Monte Carlo terminal classes.
Monte Carlo Class Overview (Major Member Functions)
Class ccMCTestable
This class provides the base functionality for Monte Carlo testing. It provides the interface (framework) for Monte Carlo. It contains only the member functions required to perform basic Monte Carlo testing. As such its primary use is as an interface specification so that Monte Carlo objects can be treated in a uniform manner.
The primary functions defined for this class are:
mcActivate(. . . ): This function performs the activation of a Monte Carlo object. Activation is used to allocate the data that a Monte Carlo object utilizes. This is done to keep the size of an unused Monte Carlo object as small as possible.
mcDeactivate(): This function deactivates an activated Monte Carlo object. It frees the data allocated by the activate.
mcReset(): This function resets a Monte Carlo object without changing the "programmed" state of the object (a discussion of the "programmed" state of a Monte Carlo object will come later in this text).
mcRunMany(. . . ): Provides the interface for running any type of Monte Carlo object. When a Monte Carlo object is activated, it is placed on a list (maintained by this class). Typically mcRunMany() would iterate over this list and run each Monte Carlo object.
mcReportSummaryAll(. . . ): Generates a summary report about the run Monte Carlo objects.
mcReportAll(. . . ): Generates a more detailed report about the run of Monte Carlo objects.
There are other member function of this class. These additional member functions are used to adjust information that the base class tracks (i.e. Monte Carlo test name, counters, timers and time statistics).
Class ccMCInspection
This class provides the interface for classes that are derived from the SMD PGP inspection class hierarchy. Since this class is used to help test SMD PGP inspections it adds additional information to the Monte Carlo object. Specifically it adds support for images (used to generate synthetic devices, fiducials, etc.), translation and rotation limits, centers, statistics about the inspections performance and the summary and more detailed reports. The primary functions introduced in this class are:
mcSetNominalCtr(. . . ): This function allow a user access to a nominal center object. It is from this nominal center object that random centers are generated. A center object defines the translation from physical space (the space were the device exists) to image space (the space where the imaging system "sees" the device).
mcSetCtrLimits(. . . ): This function places limits of the range of values that a center object can be varied by. This effectively moves the center object around.
mcSetLimits(. . . ): This function places limits of the range of values that the synthetically created object to be inspected can be varied by. It represents the variations in pickup positions by the robotic assembly equipment.
mcSetLimitSelect(. . . ): This function selects the type of variation that the limits will use. For example one could set the limit select value to random. Using this value the Monte Carlo test object would be randomly positioned and rotated within the image. Other possible values would be to use the minimum, maximum, nominal or other selection criterion.
mcSetBackgroundImage(. . . ): This function allows the user to provide an arbitrary background image on which to draw the synthetic image.
mcSetBackgroundColor(. . . ): This function allows the user to set the background color (grey value) range on which to draw the synthetic image.
mcGetDeviation(. . . ): This function get the deviation between the expected and actual inspection. This is one of the primary values that users will use when running Monte Carlo tests. A large deviation generally indicates that the inspection did a poor job.
mcSetDeviationThreshold(. . . ): This function allows the user to set the allowable deviation before signaling an alert.
There are other member functions that are used to generate and report on statistics about the expected and actual results of an inspection. Additionally, the each set or get function has a get or set function (respectively).
Class ccMCDeviceInspection
This class provides the interface for the classes that are derived from the SMD PGP ccDeviceInspection class. This class adds specific support for device inspection. The primary members introduced at this point are:
mcConstructDev2Img(): Constructs a dev2img linear transform that is used to draw the synthetic device. A dev2img is the transform that is used to go from device space (a space that is aligned with physical space and has the same scale) to image space. This transform is used to convert coordinates on a device into coordinates in an image (pixel locations).
mcDrawSyntheticDevice(. . . ): This is the function that draws the synthetic device. At this level in the hierarchy this is a pure virtual function since we do not yet know what type of device needs to be drawn. The drawing will occur at the coordinates indicated by the dev2img transform that was constructed by the above member function.
mcFromMajorAxis(. . . ): Used to select randomized presentation angles. The randomized angle is combined with the initial dev2phys (which is the 3 parameter (rigid body x, y, theta!) transform which specifies the nominal position of the device on the robotic assembly systems pickup.
mcSetInitialDev2PhysLimits(. . . ): Sets the initial dev2phys limits which will be used to orient the device. The initial dev2phys is a transform that places the origin of the physical coordinate system into device space (where the device is described).
mcRun(. . . ): This is the main execution function for a Monte Carlo run. This mcRun() is responsible for setting up the dev2img, tracking statistics and calling the inspection.
The other member functions of this class are used to record statistics on the presentation position and the measured device location, generate detailed reports and generate the randomized drawing positions.
Class ccMCRectilinearInspection
This class provides the interface for the classes that are derived from the SMD PGP ccRectDeviceInspection class. The ccRectDeviceInspection class is used for devices that can be described using the rectilinear device description (which describes a devices that have a rectilinear attributes). This class adds specific support for drawing rectilinear devices. The primary members added at this point are:
mcSet???Selector(. . . ): Selects how values should be generated for a Monte Carlo generation. For example you can select that dimensions should be the minimum, maximum, nominal or random (between min and max). Note: The ??? represent the different attributes that can be selected, such as, body size, lead size, lead color, etc.
mcSet???Range(. . . ): Sets the valid range for various rectilinear device attributes such as lead color and lead bend angle. These values are not needed by the inspection code so they are not available in the cdd.sub.-- rectilinear.sub.-- device descriptions.
Other Monte Carlo Classes
The hierarchy diagram of FIG. 4 shows additional Monte Carlo classes. These classes are provided as examples of the types of classes that can be derived from the existing Monte Carlo classes. These classes would contain data and functionality that would be appropriate for their types. However, in general, they would provide similar functionality as in the ccMCRectilinearInspection class.
Monte Carlo Helpers
In order to implement the Monte Carlo classes, several additional helper classes have been developed. These classes provide statistics gathering, random number generation, ranges (or limits). Below is a description of each of these helper classes:
class ccStatisticsMC: This class is a statistics class. Besides supporting various statistics (min, max, mean, sd, n), it also keeps absolute value statistics. That is the input value has it's absolute value taken prior to accumulating into the statistics.
class ccRandomUniform: This class provides support for generating uniform random numbers. The generator can be reseeded and it's state can be saved so that it may be reloaded at a later time (to provide a "replay" capability).
Other Support Functions
Other support functions have been developed to assist in using the Monte Carlo classes. The main support function is the anti-aliased drawing functions. These routines are used to generate synthetic data which is then typically used to perform an inspection on. It is important that these drawings are as close to a real object as it is seen by the camera so that the results of the Monte Carlo test is not biased. Of course it isn't possible for a synthetically generated device to look identical to a real device. Using anti-aliased drawings allows for a reasonably close approximation. Without such drawings, inspection might fail or produce invalid results. Basically, the anti-aliased drawings allow for the generation of "perfect" devices with "perfect" defects. It is up to the class developer as to how much realism is permitted and needed when generating synthetic objects. A description of the anti-aliased drawing function is not provided. Numerous examples can be found in the literature for graphics rendering.
Monte Carlo Goals
Low Impact on Other Classes
Monte Carlo is intended to be used as a mixing class (multiple inheritance or MI) for other classes that one might wish to test. It was a design goal that Monte Carlo did not impose itself into the design of other classes. This means that a developer can provide a Monte Carlo version that is separate from the non-Monte Carlo version and the user can pick which version to use. For example, the class layout shown in FIG. 5 can exist within the machine vision inspection system.
In this case one might wish to provide a version of ccLargeLeadedInspection that is also Monte Carlo enabled. Using MI, one would create a class layout of the type shown in FIG. 6. Now a user who wishes to have a Monte Carlo large leaded inspection could do so and it would not impact the user who does not wish to have such an inspection.
It should be noted that the MI design does not prevent a developer from providing only a Monte Carlo enabled inspection. If such a design is desired, the developer would change the class inheritance so that ccLargeLeadedInspection is derived from ccRectDeviceInspection and from ccMCRectilinearInspection. This might be desirable if a developer does not want to support multiple instances of an inspection. Because the non-activated portion of a Monte Carlo enabled inspection requires little runtime memory, this design choice is quite reasonable.
Flexibility of Implementations
The class layout of Monte Carlo is designed to decrease the amount of work that an developer needs to do to add in Monte Carlo. It is also designed to be flexible so that a developer of the ccMCDeviceInspection Monte Carlo class doesn't have to live with the constraints that a developer of the ccMCFiducialInspection class imposes. In addition, if a particular class doesn't do all the required operations, the developer can override much of the implementations or provide additional member functions in a new derived class. For example if a developer of the ccMCRectilinearInspection class wanted to add in the ability of specify printing that should appear on the synthetically created device new member functions could be added to a new class (this new class could be a terminal inspection/test class or an additional test class) that could be used to specify the printing that should appear. Another example would be to provide for backlight inspections. A developer could override the background color and device color member functions and provide versions that implement backlighting.
Low Memory Usage
Since it is expected that users will find Monte Carlo capability useful in a number of situations it is expected that Monte Carlo based version of many objects might exist in the system at once. The Monte Carlo portions of objects might be very large due to the additional information that Monte Carlo requires (selectors, colors, statistics, termination criteria, etc.). Since these objects will normally not be running in Monte Carlo mode it was desirable to be able to reduce the memory foot print of the Monte Carlo enabled object. This is accomplished via the mcActivate() and mcDeactivate() member functions. Prior to calling mcActivate() a Monte Carlo object should be in a state that minimizes data storage requirements. This is typically implemented using a pointer to some private class or structure that will be null when a Monte Carlo enabled object is constructed. Calling the mcActivate() will allocate the memory that is needed to store any and all information that a Monte Carlo object must manipulate. After completion of a Monte Carlo sequence the user can call mcDeactivate() which will free the memory that was allocated by the mcActivate(). In this way a Monte Carlo's memory footprint can be kept low when the object is not being used in a Monte Carlo test.
Low Cost to Monte Carlo Enable an Object
It is desirable for one to be able to use Monte Carlo without having to do anything to ones code. While this is the desire, in practice one typically needs to do something to add in Monte Carlo support. The goal here was to make this additional effort as small as possible. As part of this goal, wherever possible, the Monte Carlo object should do some reasonable default behavior so that the user (both a class developer and a class user) can start using Monte Carlo as quickly as possible. Here is the default implementation for mcDeviation().
While this implementation might not be the best representation of the deviation of the device it provides a reasonable approximation. A class developer could use this version to start initial testing and later override it to provide a more appropriate implementation.
Additionally, a user might create an object that is Monte Carlo enabled, activate that object and then run the object. Here the object might default things like lead color, selectors, etc. so that the user doesn't need specify these values to start running a reasonable Monte Carlo run.
When a user of the terminal Monte Carlo classes mixes in Monte Carlo support they typically only need to implement a few member functions. For example, to implement a Monte Carlo class based on the ccMCRectilinearInspection class, one needs to provide the following member functions:
mcGetInspection(): Return the this pointer so that the Monte Carlo class can access the inspection base class.
mcGetUncertainty(): Return the device uncertainty.
In general, these member functions are very simple member functions to implement.
If one wishes to use a non-terminal class then more work is required. For example to implement a new class off the ccMCInspection class, it is necessary to that the developer provides at least the following member functions:
mcRun(): Implements the run method for this object. Typically this function is responsible for testing termination criteria, drawing the device, performing the inspection operation and recording statistics.
mcGetUncertainty(): Return the device uncertainty.
mcGetBoundingBox(): Return the bounding box of the object to be tested.
mcGetInspection(): Return the this pointer so that the Monte Carlo class can access the inspection base class.
While most of these member functions are simple to implement, the mcRun() member function is one of the hardest to write. It is responsible for testing termination criteria, drawing the device (which might be very difficult to implement), etc. But even at this level in the hierarchy one does not need to implement many of the other support functions in Monte Carlo (report generation, mcRunMany(), timing functions, random number generation, etc.
Monte Carlo Implementation
The four classes shown in the Appendix are used to implement Monte Carlo. The first is the Monte Carlo base class ccMCTestable. It is this class that defines the primary public interface for a Monte Carlo object.
The second class provides a sample of a more derived implementation. This class adds additional termination criteria and implements several important member functions that could not be implemented in the base class. Note the class ccCenter is used by the ccInspection class hierarchy and is not a part of the Monte Carlo class design.
The third class provides additional support for generating Monte Carlo test runs. It provides an interface for drawing devices on an image. Note that the class cc2DMap is a 2 dimension transform class used to define transforms between physical units of measure and pixel units and is not a part of the Monte Carlo class design.
The forth class is a terminal Monte Carlo class. It's primary purpose is to provide drawing support as well as setter/getters for various drawing attributes that might not be contained in the device description that it gets from the ccRectDeviceInspection class.
Described above are machine vision methods meeting the objects set forth. These methods provide improved methods for testing machine vision inspection systems overcoming the drawbacks of the prior art techniques. It will be appreciated that the embodiments described above are illustrative only and that additional embodiments within the ken of those of ordinary skill in the art fall within the scope of the invention. ##SPC1##
Claims
In view of the foregoing, what I claim is:
1. A method of testing a machine vision inspection system of the type that inspects a feature in an image using digital data processing object-oriented constructs that instantiate an inspection object for the feature from an inspection class associated with a type of the feature; invoke a method member of that inspection object to inspect the feature to determine a characteristic thereof, that method member being referred to herein as an inspection method, the method comprising the steps of: instantiating a test object from a test class that inherits from the inspection class; invoking a method member of the test object for generating one or more test images representing the feature; invoking the inspection method to inspect the test images to determine characteristics of the features therein; and reporting a result of such inspection.
2. A method according to claim 1, wherein the step of generating one or more test images includes the step of generating the test image representing the feature with a characteristic having a value selected from multiple potential values.
3. A method according to claim 2, wherein the step of generating one or more test images includes the step of generating the test image representing the feature with a characteristic having a value selected from multiple potential values, the characteristic being any of position, angular orientation, size, and color.
4. A method according to claim 2, wherein the step of generating one or more test images includes the step of generating the test image representing the feature with a characteristic having a value that is selected randomly from among multiple potential values.
5. A method according to claim 4, wherein the step of generating one or more test images includes the step of generating the test image representing the feature with multiple characteristics, each having a value which is selected randomly from among multiple potential values.
6. A method according to claim 2, wherein the step of generating one or more test images includes generating multiple test images, each representing the feature with a characteristic having a value that is selected randomly from among multiple potential values.
7. A method according to claim 6, wherein the step of generating one or more test images includes generating multiple test images, each representing the feature having multiple characteristics, each of which has a value that is selected randomly from among multiple potential values.
8. A method according to claim 1, wherein the step of generating one or more test images includes the step of generating multiple test images, each representing the feature with a characteristic having a value that is fixed.
9. A method according to claim 1, wherein the step of generating one or more test images includes the step of generating multiple test images, each representing the feature with a first characteristic having a value that is selected in random from among multiple potential values, and having a second characteristic having a value that is fixed.
10. A method according to claim 9, wherein the step of generating multiple test images includes the steps of selecting as the first characteristic a characteristic from the set of characteristics including position, angular orientation, size, and color; and selecting as the second characteristic a different characteristic from that set.
11. A method for testing a machine vision inspection system of the type that inspects a feature in an image using digital data processing object-oriented constructs that instantiate an inspection object for the feature from an inspection class associated with a type of the feature; invoke a method member of that inspection object to inspect the feature to determine a characteristic thereof, that method member being referred to herein as an inspection method, the method comprising the steps of: instantiating a test object from a test class corresponding to the inspection class, the test class inheriting from that inspection class, as well as from a test harness class; invoking a method member of the test object for generating one or more test images representing the feature; invoking the inspection method of that inspection object to inspect the test images to determine characteristic of the features therein; and reporting a result of such inspection.
12. A method according to claim 11, wherein the step of instantiating the inspection object includes the step of instantiating that object from the test class.
13. A method according to claim 11, wherein the step of reporting a result of the inspection includes the step of using members of the test object inherited from the test harness class to report a result of the inspection.
14. A method according to claim 11, wherein the test class inherits from the inspection class a definition of one or more characteristics of the feature and expected values thereof, and wherein the step of generating one or more test images includes the step of generating a test image representing the feature with a characteristic having a value selected from any of the expected values defined by the inspection class.
15. A method according to claim 14, wherein the step of generating one or more test images includes the step of generating the test image representing the feature with a characteristic having a value that is at least one of fixed, selected randomly from among the expected values defined by the inspection class, and selected in a predetermined sequence from among the expected values defined by the inspection class.
16. A method according to claim 14, wherein the step of generating one or more test images includes generating multiple test images, each representing the feature with a characteristic having a value that is at least one of fixed, selected randomly from among the expected values defined by the inspection class, and selected in a predetermined sequence from among the expected values defined by the inspection class.
17. A method for testing a machine vision inspection system of the type that inspects a feature in an image using digital data processing object-oriented constructs that instantiate an inspection object for the feature from an inspection class associated with a type of the feature; invoke a method member of that inspection object to inspect the feature to determine a characteristic thereof, that method member being referred to herein as an inspection method, the method comprising the steps of: instantiating a test object from a test class corresponding to the inspection class, the test class inheriting from that inspection class, as well as from a test harness class; invoking a method member of the test object for generating multiple test images of the feature, each test image representing a feature with a characteristic having a value selected from multiple expected values defined by the inspection class; invoking the inspection method of the inspection object to inspect the test images to determine characteristics of the features therein; comparing results of those inspections with results expected in accord with the selection of values, and tracking results of those comparisons to evaluate the inspection model.
18. A method generating images for testing a machine vision inspection system of the type that inspects a feature in an image using digital data processing object-oriented constructs that instantiate an inspection object for the feature from an inspection class associated with a type of the feature; invoke a method member of that inspection object to inspect the feature to determine a characteristic thereof, that method member being referred to herein as an inspection method, the method comprising the steps of: instantiating a test object from a test class that inherits from the inspection class; invoking a method member of the test object for generating one or more test images of the feature; invoking the inspection method to inspect the test images to determine characteristics of the features therein.
19. A method for generating images for testing a machine vision inspection system of the type that inspects a feature in an image using digital data processing object-oriented constructs that instantiate an inspection object for the feature from an inspection class associated with a type of the feature; invoke a method member of that inspection object to inspect the feature to determine a characteristic thereof, that method member being referred to herein as an inspection method, the method comprising the steps of: instantiating a test object from a test class corresponding to the inspection class, the test class inheriting from that inspection class, as well as from a test harness class; invoking a method member of the test object for generating multiple test images of the feature, each test image representing the feature with a characteristic having a value selected from multiple expected values defined by the inspection class; invoking the inspection method of the inspection object to inspect the test images to determine characteristics of the features therein.
20. A method for testing a machine vision surface mounted device inspection system of the type that inspects a surface mounted device in an image using digital data processing object-oriented constructs that instantiate an inspection object for the surface mounted device from an inspection class associated with a type of the surface mounted device; invoke a method member of that inspection object to inspect the surface mounted device to determine a characteristic thereof, that method member being referred to herein as an inspection method, the method comprising the steps of: instantiating a test object from a test class corresponding to the inspection class; invoking a method member of the test object for generating one or more test images of the surface mounted device; invoking its inspection method to inspect each of the test images to determine a characteristic of the surface mounted devices therein; and reporting a result of such inspection.
Patent Citations (134)
| Patent | Date | Inventor | Cited By |
|---|---|---|---|
| US3816722 | 1974-06-01 | Sakoe et al. | |
| US3936800 | 1976-02-01 | Ejiri et al. | |
| US3967100 | 1976-06-01 | Shimomura | |
| US3968475 | 1976-07-01 | McMahon | |
| US3978326 | 1976-08-01 | Shimomura | |
| US4011403 | 1977-03-01 | Epstein et al. | |
| US4115702 | 1978-09-01 | Nopper | |
| US4115762 | 1978-09-01 | Akiyama et al. | |
| US4183103 | 1980-01-01 | Agrawala et al. | |
| US4200861 | 1980-04-01 | Hubach et al. | |
| US4254400 | 1981-03-01 | Yoda et al. | |
| US4300164 | 1981-11-01 | Sacks | |
| US4385322 | 1983-05-01 | Hubach et al. | |
| US4441124 | 1984-04-01 | Heebner et al. | |
| US4441206 | 1984-04-01 | Kuniyoshi et al. | |
| US4534813 | 1985-08-01 | Williamson et al. | |
| US4541116 | 1985-09-01 | Lougheed | |
| US4570180 | 1986-02-01 | Baier et al. | |
| US4577344 | 1986-03-01 | Warren et al. | |
| US4581762 | 1986-04-01 | Lapidus et al. | |
| US4606065 | 1986-08-01 | Beg et al. | |
| US4617619 | 1986-10-01 | Gehly | |
| US4630306 | 1986-12-01 | West et al. | |
| US4688088 | 1987-08-01 | Hamazaki et al. | |
| US4706168 | 1987-11-01 | Weisner | |
| US4728195 | 1988-03-01 | Silver | |
| US4730260 | 1988-03-01 | Mori et al. | |
| US4731858 | 1988-03-01 | Grasmueller et al. | |
| US4736437 | 1988-04-01 | Sacks et al. | |
| US4742551 | 1988-05-01 | Deering | |
| US4764870 | 1988-08-01 | Haskin | |
| US4771469 | 1988-09-01 | Wittenburg | |
| US4783826 | 1988-11-01 | Koso | |
| US4783828 | 1988-11-01 | Sadjadi | |
| US4783829 | 1988-11-01 | Miyakawa et al. | |
| US4831580 | 1989-05-01 | Yamada | |
| US4860374 | 1989-08-01 | Murakami et al. | |
| US4860375 | 1989-08-01 | McCubbrey et al. | |
| US4876457 | 1989-10-01 | Bose | |
| US4876728 | 1989-10-01 | Roth | |
| US4903218 | 1990-02-01 | Longo et al. | |
| US4907169 | 1990-03-01 | Lovoi | |
| US4914553 | 1990-04-01 | Hamada et al. | |
| US4922543 | 1990-05-01 | Ahlbom et al. | |
| US4926492 | 1990-05-01 | Tanaka et al. | |
| US4932065 | 1990-06-01 | Feldgajer | |
| US4953224 | 1990-08-01 | Ichinose et al. | |
| US4955062 | 1990-09-01 | Terui | |
| US4959898 | 1990-10-01 | Landman et al. | |
| US4962423 | 1990-10-01 | Yamada et al. | |
| US4972359 | 1990-11-01 | Silver et al. | |
| US4982438 | 1991-01-01 | Usami et al. | |
| US5012402 | 1991-04-01 | Akiyama | |
| US5046190 | 1991-09-01 | Daniel et al. | |
| US5054096 | 1991-10-01 | Beizer | |
| US5060276 | 1991-10-01 | Morris et al. | |
| US5063608 | 1991-11-01 | Siegel | |
| US5073958 | 1991-12-01 | Imme | |
| US5081656 | 1992-01-01 | Baker et al. | |
| US5081689 | 1992-01-01 | Meyer et al. | |
| US5086478 | 1992-02-01 | Kelly-Mahaffey et al. | |
| US5090576 | 1992-02-01 | Menten | |
| US5091861 | 1992-02-01 | Geller et al. | |
| US5091968 | 1992-02-01 | Higgins et al. | |
| US5093867 | 1992-03-01 | Hori et al. | |
| US5113565 | 1992-05-01 | Cipolla et al. | |
| US5115309 | 1992-05-01 | Hang | |
| US5119435 | 1992-06-01 | Berkin | |
| US5124622 | 1992-06-01 | Kawamura et al. | |
| US5133022 | 1992-07-01 | Weideman | |
| US5134575 | 1992-07-01 | Takagi | |
| US5143436 | 1992-09-01 | Baylor et al. | |
| US5145432 | 1992-09-01 | Midland et al. | |
| US5151951 | 1992-09-01 | Ueda et al. | |
| US5153925 | 1992-10-01 | Tanioka et al. | |
| US5159281 | 1992-10-01 | Hedstrom et al. | |
| US5159645 | 1992-10-01 | Kumagai | |
| US5164994 | 1992-11-01 | Bushroe | |
| US5168269 | 1992-12-01 | Harlan | |
| US5185855 | 1993-02-01 | Kato et al. | |
| US5189712 | 1993-02-01 | Kajiwara et al. | |
| US5206820 | 1993-04-01 | Ammann et al. | |
| US5216503 | 1993-06-01 | Paik | |
| US5225940 | 1993-07-01 | Ishii et al. | |
| US5230027 | 1993-07-01 | Kikuchi | |
| US5243607 | 1993-09-01 | Masson et al. | |
| US5253306 | 1993-10-01 | Nishio | |
| US5253308 | 1993-10-01 | Johnson | |
| US5265173 | 1993-11-01 | Griffin et al. | |
| US5271068 | 1993-12-01 | Ueda et al. | |
| US5287449 | 1994-02-01 | Kojima | |
| US5297256 | 1994-03-01 | Wolstenholme et al. | |
| US5299269 | 1994-03-01 | Gaborski et al. | |
| US5311598 | 1994-05-01 | Bose et al. | |
| US5315388 | 1994-05-01 | Shen et al. | |
| US5319457 | 1994-06-01 | Nakahashi et al. | |
| US5327156 | 1994-07-01 | Masukane et al. | |
| US5337262 | 1994-08-01 | Luthi et al. | |
| US5337267 | 1994-08-01 | Colavin | |
| US5363507 | 1994-11-01 | Nakayama et al. | |
| US5367439 | 1994-11-01 | Mayer et al. | |
| US5367667 | 1994-11-01 | Wahlquist et al. | |
| US5371690 | 1994-12-01 | Engel et al. | |
| US5388197 | 1995-02-01 | Rayner | |
| US5388252 | 1995-02-01 | Dreste et al. | |
| US5398292 | 1995-03-01 | Aoyama | |
| US5432525 | 1995-07-01 | Maruo et al. | |
| US5440699 | 1995-08-01 | Farrand et al. | |
| US5455870 | 1995-10-01 | Sepai et al. | |
| US5455933 | 1995-10-01 | Schieve et al. | |
| US5475766 | 1995-12-01 | Tsuchiya et al. | |
| US5477138 | 1995-12-01 | Erjavic et al. | |
| US5481712 | 1996-01-01 | Silver et al. | |
| US5485570 | 1996-01-01 | Bushboom et al. | |
| US5491780 | 1996-02-01 | Fyles et al. | |
| US5495424 | 1996-02-01 | Tokura | |
| US5495537 | 1996-02-01 | Bedrosian et al. | |
| US5519840 | 1996-05-01 | Matias et al. | |
| US5526050 | 1996-06-01 | King et al. | |
| US5532739 | 1996-07-01 | Garakani et al. | |
| US5550763 | 1996-08-01 | Michael | |
| US5566877 | 1996-10-01 | McCormack | |
| US5568563 | 1996-10-01 | Tanaka et al. | |
| US5574668 | 1996-11-01 | Beaty | |
| US5574801 | 1996-11-01 | Collet-Beillon | |
| US5583949 | 1996-12-01 | Smith et al. | |
| US5583954 | 1996-12-01 | Garakani | |
| US5592562 | 1997-01-01 | Rooks | |
| US5594859 | 1997-01-01 | Palmer et al. | |
| US5602937 | 1997-02-01 | Bedrosian et al. | |
| US5608872 | 1997-03-01 | Schwartz et al. | |
| US5640199 | 1997-06-01 | Garakani et al. | |
| US5640200 | 1997-06-01 | Michael | |
| US5715369 | 1998-02-01 | Spoltman et al. |
Non-Patent Literature (13)
- Grimson, W. Eric L. and Huttenlocher, Daniel P., "On the Sensitivity of the Hough Transform for Object Recognition", May 1990, IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 12, No. 3.
- Medina-Mora et al. (1981) An Incremental Programming Environment, IEEE Transactions on Software Eng. SE-7:472-482.
- Teitelbaum et al. (19810 The Cornell Program Synthesizer: A Syntax-Directed Programming Environment, Communications of the ACM 24:563-573.
- Newsletter from Acquity Imaging, Inc., "Remote Vision Support Package--The Phones Are Ringing|," 1 page.
- PictureTel Corporation Product Brochure "Picturetel Live PCS 100(tm) Personal Visual Communications System," 3 pp. (1993).
- PictureTel Corporation Product Brochure "Picturetel System 1000: Complete Videoconferencing for Cost Sensitive Applications," 4 pp. (1993).
- PictureTel Corporation Product Brochure, "Picturetel System 4000(tm) A Family of Models to Fit Your Application From Offices to Boardrooms, Classrooms, and Auditoriums," 4 pp. (1993).
- Symantec Corporation, "The Norton pcAnywhere User's Guide," Table of Contents 8 pp; Introduction of pcAnywhere Technology pp. i-vii; Chapter 7-Sessions; pp. 191-240 (1991).
- Bursky, Dave, "CMOS Four-Chip Set Process Images at 20-MHz Data Rates," Electronic Design, May 28, 1987, pp. 39-44.
- Plessey Semiconductors, Preliminary Information, May 1986, Publication No. PS2067, May 1986, pp. 1-5.
- NEC Electronics Inc., PD7281 Image Pipelined Processor, Product Information Brochure, pp. 2-169-2-211.
- Horn, Berthold Klaus Paul. "Robot Vision", The Massachusetts Institute for Technology, 1986.
- Rosenfeld, Azriel. "Computer Vision: Basic Principles," Proceedings of the IEEE. vol. 76, No. 8, Aug. 1988. pp. 863-868.