US 5,289,993 AGrant
Method and Apparatus for Tracking an Aimpoint With Arbitrary Subimages
Issue Date:1994-03-01
•18 Claims
•7 Drawing Sheets
Abstract
A method for tracking an aimpoint with arbitrary subimages as described. The method comprises the steps of acquiring an aimpoint (50) on a target and a set of subimages (80) arbitrarily associated with the aimpoint by a sensor in a field of view. For a first time, the dimensionality of each subimage (82) and a normalized distance (54) from each subimage to the aimpoint are determined. For a second time, at least one of the subimages is reacquired (56) and the subsequent location of the aimpoint is estimated (58) based on the subsequent location of the subimages, on the normalized distances, and, where a subimage is one-dimensional, on a maximum eigenvector associated with the subimage.
Metadata
Inventors
- Joel K. McWilliams
- Don R. Van Rheeden
Application Information
Application Number:US 954643&
Filing Date:1992-09-30
Priority Date:1991-08-30
Art Unit:221
Classifications
IPC:
F41G 730
Field of Search:
2443823.15;3.16;3.171;48
Patent Drawings (7 sheets)
Description
TECHNICAL FIELD OF THE INVENTION
This invention relates to computer control and more particularly to a method of tracking an aimpoint with arbitrary subimages.
BACKGROUND OF THE INVENTION
In certain computer control applications, it is necessary to track and measure the image of an object passively. It is especially important in weapons delivery systems that a target be so tracked. If such a target were tracked actively (i.e., using radar or laser range finding techniques), the target might detect the presence of the tracker. Once the target has detected the presence of the tracker, it can respond in one of several ways, all of which are deleterious to the tracker. For instance, the target might "jam" the tracker by bombarding it with signals that are comparable to those which the tracker is actively using or the target might fire its own weapon at the tracker, at the source of the tracking signal, or, even at the launching site of the tracker. In this way, the target could defeat the tracker, destroy the tracker or perhaps even destroy the launch site of the tracker, including the operating personnel.
Passively tracking a target, however, imposes at least one serious limitation on the tracker. A tracker cannot accurately determine the distance or "range" to a target if it cannot actively sense the object. An active tracker, for instance, could determine the distance to a target by measuring the elapsed time from the emission of a radio frequency signal to the receipt of the signal reflected off of the target. The absence of a range measurement from tracker to target limits the passive tracker's ability to compensate for the apparent change in target image as the tracker moves in relationship to the target. Without this ability, a tracker will fail to maintain a constant target aimpoint.
In practice, a tracker benefits by tracking several subimages of its target's image. These subimages are two dimensional representations of structures that are physically related to the exact target location or "aimpoint" in the real three-dimensional world. Multiple subimages are used for redundancy purposes and because the actual aimpoint of the target is often untrackable due to low image contrast, brightness, or other reasons. As the tracker nears the target, however, the subimages will appear to radiate outwardly with respect to each other. The position of the subimages with respect to one another may also change in other ways in certain situations. For instance, two subimages located on a target may appear to approach one another if they are located on a face of a target that is rotating away from the tracker. A subimage associated with a straight edged structure or a line poses other problems for a passive tracker. Such a subimage is indistinguishable from other subimages corresponding to other points on the line. This ambiguity can introduce serious error into a track. Straight edges typically appear when tracking structures such as buildings, bridges, dams or internal detail on armored vehicles. The tracker motion can be further complicated by background subimages erroneously tracked by the tracker. A tracker will then mislocate the aimpoint and, perhaps, completely miss its target if it cannot identify and compensate for "bad" subimages.
Prior attempts to passively track an object have resulted in solutions with limited flexibility and poor accuracy. Heretofore, an object once identified as an aimpoint was tracked by tracking a predetermined number of subimages in a known pattern. Typically, the pattern chosen was a square with the aimpoint at its center and four subimages located at the four corners of the square. That system would track the four subimages located at the corners of the square and infer the actual aimpoint using the simple symmetry of the predetermined square. This method faltered when the geometry of the actual target resulted in less than four suitable subimages located in the requisite pattern or when the subimage selected was not associated with the aimpoint. This system lacked the ability to identify bad subimages.
Therefore, a need has arisen for a passive subimage tracker which is able to identify trackable subimages in arbitrary image locations and use them to track an aimpoint.
SUMMARY OF THE INVENTION
In accordance with the present invention, a method and apparatus for tracking an aimpoint with arbitrary subimages is provided which substantially eliminates or reduces disadvantages and problems associated with prior trackers.
A method for tracking an aimpoint with arbitrary subimages is described. The method comprises the steps of acquiring an aimpoint on a target and a set of subimages arbitrarily associated with the aimpoint by a sensor in a field of view. For a first time, the dimensionality of each subimage and a normalized distance from each subimage to the aimpoint are determined. For a second time, at least one of the subimages is reacquired and the subsequent location of the aimpoint is estimated based on the subsequent location of the subimages and on the normalized distances. Where a subimage is determined to be one-dimensional, a maximum eigenvector is calculated for the subimage and the component of the normalized distance parallel to the eigenvector is used.
It is a first technical advantage of the invention that the aimpoint tracker can detect and use subimages associated with lines as well as two-dimensionally distinct subimages. This improves accuracy twice. First, each subimage is modeled with the appropriate (one- or two-dimensional) mathematical model. Second, more subimages are available to track an aimpoint. Each consequence advantageously affects tracker accuracy.
A second technical advantage of the invention is the tracker's ability to reject subimages which yield poor results. Typically, these subimages yield such results because they represent image clutter which is not physically associated with the target. This improves the future confidence of aimpoint calculations.
A third technical advantage of the invention is its enhanced ability to redetermine the aimpoint based on an adjusted set of subimages. The adjusted set of subimages includes only those subimages which behave as though they are connected to the aimpoint in the real world. This improves the immediate confidence of aimpoint calculations and the accuracy of the device, such as a missile, whose movement path is charted based on the perceived aimpoint.
BRIEF DESCRIPTION OF THE DRAWINGS
For a more complete understanding of the present invention, and the advantages thereof, reference is now made to the following descriptions taken in conjunction with the accompanying drawings, in which:
FIG. 1 is a perspective schematic view of the use of an anti-armor "fire and forget" missile system which may advantageously incorporate the disclosed invention;
FIG. 2 is a schematic, part cross-sectional diagram of the anti-armor missile of FIG. 1;
FIG. 3 is a half-tone replication of a digitized image of the target at a certain time as seen by the "fire and forget" missile depicted in FIG. 1;
FIG. 4 depicts trackable subimages associated with the target depicted in FIG. 1 at the time corresponding to the time of FIG. 3;
FIGS. 5 and 6 are schematic elevational views depicting the image and trackable subimages of the target from FIG. 1 at a first time and at a subsequent time, respectively;
FIG. 7 depicts a flow chart of a subimage tracker according to one embodiment of the disclosed invention;
FIG. 8 depicts a target being tracked by four subimages where one subimage does not move with the aimpoint;
FIG. 9 mathematically depicts the target of FIG. 3 in a two-dimensional coordinate system;
FIG. 10 depicts a flow chart of the disclosed general method of the invention;
FIG. 11 depicts a target having at least one subimage associated with a line;
FIGS. 12a and 12b graphically depict a correlator test associated with a two-dimensional and a one-dimensional subimage, respectively;
FIGS. 13a and 13b graphically depict the surfaces generated by the correlator tests of FIGS. 12a and 12b, respectively;
FIG. 14 graphically depicts the orientation of correlator points used in conjunction with the surfaces of FIGS. 13a and 13b;
FIG. 15 depicts a matrix of constants used in conjunction with the correlator points of FIG. 14;
FIG. 16 mathematically depicts a one-dimensional subimage associated with a line with a first and second coordinate system;
FIG. 17 further depicts the one-dimensional subimage depicted in FIG. 16; and
FIG. 18 depicts part of a flow chart of an embodiment of the invention which distinguishes one- and two-dimensional subimages.
DETAILED DESCRIPTION OF THE INVENTION
The preferred embodiment of the present invention and its advantages are best understood by referring to FIGS. 1-18 of the drawings, like numerals being used for like and corresponding parts of the various drawings.
FIG. 1 depicts an anti-armor "fire and forget" missile system which incorporates the disclosed invention. A missile 10 is fired at a target 12 by an infantryman 14. Infantryman 14 selects an aimpoint on the target 12 that he believes will offer the least resistance to missile 10 upon impact. Missile 10 incorporates a passive imaging sensor 16 and a processor (not shown) for acquiring and digitizing an image. Missile 10 also has movable fins 18 which are controlled by the processor to vary the direction of the missile. Missile 10 acquires trackable subimages related to the particular aimpoint selected by infantryman 14 which are within the field of view of the sensor 16. Missile 10 computes an initial geometric relationship between the aimpoint and the initial subimages and retains this in an onboard memory (not shown). Missile 10 periodically reacquires the subimages and computes a new aimpoint as will be more fully described below. Missile 10 may then adjust its path to intercept target 12 despite subsequent movement of the target 12 relative to infantryman 14 or missile 10.
The type of missile depicted in FIG. 1 is known as a "fire and forget" missile because infantryman 14 may forget the missile after launch and seek cover or engage other targets of opportunity. Missile 10, however, will continue to track and intercept target 12. Missile 10 may select subimages based on any number of criteria including, but not limited to, the "hot spot" and "correlator" methods. In the "hot spot" method, a sensor identifies subimages according to the hottest (brightest or warmest) spot within a specified region. In the correlator method, the tracker searches for a previously identified subimage.
FIG. 2 depicts "fire and forget" missile 10 which may incorporate the disclosed invention. Missile 10 delivers a warhead 20 which detonates upon impact with target 12 (FIG. 1). The missile 10 contains an onboard sensor 16 such as a forward looking infrared sensor ("FLIR"), that is sensitive to radiation emitted by the missile's target. A latch 22 temporarily saves the information received by sensor 16 so that it is more accessible by a central processing unit ("CPU") 24. CPU 24 might itself comprise various subsystems (not shown) which may be implemented by hardware or software, including an aimpoint designator for initially establishing the aimpoint on the target, a normalized distance calculator for calculating the distance between each subimage and the aimpoint, and a comparator for comparing the actual and estimated subimage locations and for rejecting those subimages that fail a certain statistical test.
CPU 24 periodically acquires images within its field of view during the operation of missile 10. CPU 24 has associated with it a memory 26. Memory 26 contains the routines which CPU 24 runs and stores data necessary to the disclosed invention. CPU 24 controls the direction of missile 10 through a fin control unit 28. Fin control unit 28 manipulates each fin 18 through, for example, a servo 30. Missile 10 is propelled by a rocket motor 32.
FIG. 3 depicts a view of target 12 by sensor 16 of missile 10. The aimpoint, as originally specified by infantryman 14, is indicated by a crosshairs. In this particular embodiment, missile 10 has an infrared sensor. The engine compartment and wheel bearings of target 12, being hottest, are therefore indicated by the color white. Cooler structures such as the background are indicated by shades of grey and by black.
FIG. 4 depicts certain trackable subimages contained in FIG. 3. Each trackable subimage is identified as a small square. A tracker on board missile 10 initially selects a number of these subimages to track the target at an initial time. The tracker also calculates a normalized distance between the aimpoint selected by the infantryman 14 and each subimage. This distance may be the actual number of pixels between the aimpoint and the subimage as sensed by the tracker. The distance may be further normalized so that at least one subimage is a certain distance, such as "1", away from the aimpoint. These dimensions are saved for future use by the tracker.
FIG. 5 depicts a view of target 12 and six trackable subimages superimposed thereon. FIG. 6 depicts the view of target 12 and the trackable subimages as depicted in FIG. 5 at a slightly later time. The image of target 12 appears to grow in size as the missile 10 (FIGS. 1 and 2) closes in on target 12. As a consequence, the subimages depicted also appear to grow in size and to move radially outward from the aimpoint. Three subimages, initially at the right side of FIG. 5, are no longer visible in the field of view. The tracker will typically adjust its course to maintain the aimpoint (crosshairs) in the center of its field of view. This apparent movement of the subimages relative to the aimpoint and the loss of some subimages must be accounted for by the tracker to insure successful guidance of missile 10 to target 12.
GENERALIZED GEOMETRIC MODEL
The approach to multiple subimage tracking without range estimates is based on a generalized geometric model. This model is based on the fact that though the target and thus the subimages will be growing in the image during closure to the target, the relative dimensions of the target do not change. This assumes that the angle of attack between the tracker and target stays fairly constant, which is common during most of the terminal phase of the missile flight.
In the generalized geometric approach each tracker location is related to the aimpoint location using a normalized coordinate frame. An individual subimage i at image location (x.sub.i, y.sub.i) can be related to the aimpoint A at image location (x.sub.A, y.sub.A) by the following equations:
where (.delta.x.sub.i, .delta.y.sub.i) represents the offset in the image plane of subimage i from the aimpoint A, and (n.sub.x.sbsb.i, n.sub.y.sbsb.i) are additive noise terms which corrupt measurement of the true subimage location. These equations can be combined into a single equation using vector notation:
The key to accurately modeling each subimage position is the accurate representation of the offset vector. The offset vector may be expressed as: ##EQU1## where s is the target image magnification factor ("TIMF") and (d.sub.x.sbsb.i, d.sub.y.sbsb.i) are the size normalized distances between a subimage i and the aimpoint A. Thus (d.sub.x.sbsb.i, d.sub.y.sbsb.i) are range independent normalized distances. Although size is typically chosen as the normalization factor because it is a measurable quantity of the target, any scaling factor can be used. This model leads to the following matrix representation for a subimage i: ##EQU2## For N subimages, the previous equation may be expanded: ##EQU3## Equation (6) may be rewritten as
where
x is a vector of subimage locations and H is the 2N.times.3 matrix of "1"s, "0"s, and normalized distances depicted in Eq. (6) above. At each successive time at which the tracker reacquires the subimages, all variables are known except those in .theta. and in the noise vector n.
The vector .theta. and hence the aimpoint may be estimated by several techniques including a least squares technique where:
where .theta. is an estimate of .theta.. This method will minimize the effect of the noise vector n.
FIG. 7 depicts a flow chart of a subimage tracker that incorporates the generalized geometric approach. An aimpoint is initially acquired on the target in block 50 by an operator. The tracker then acquires multiple subimages associated with the chosen aimpoint according to internal criteria such as image contrast or image brightness in block 52. The tracker calculates the normalized distances between each subimage that it has acquired and the selected aimpoint in block 54. These distances are stored in a matrix whose particular form is shown in Eq. (6) above. In blocks 56 and 58, the tracker reacquires the subimages and estimates the location of the aimpoint from the previously calculated normalized distances using a mathematical model such as Eq. (9) described above.
If the missile is closing on the target, some of the original subimages may leave the field of view of the sensor. Meanwhile, new trackable subimages may be found as target details appear. These new subimage locations may be added to the matrix of subimage locations, typically in block 56, as needed. The tracker may then adjust sensor pointing, block 60, to maintain the aimpoint at the center of its field of view. The final three steps depicted in FIG. 7 are repeated until the missile impacts its target or the tracker otherwise ends its program. It should be understood that block 60 may comprise any number of related steps such as issuing commands to an operator to follow the aimpoint or displaying crosshairs to pinpoint the location of the aimpoint in a display. The tracker may also be mounted in a stationary environment where it simply follows an aimpoint in its field of view without actively pursuing the target.
GENERALIZED ERROR CHECKING ROUTINE
FIG. 8 is an elevational view of a targeted tank in which the tank 62 is moving from left to right with a velocity v(t). A trackable clutter object 64 in the upper left correlator (indicated by the burning oil well pump) will introduce a bias into the original aimpoint 66. If the clutter happens to be more trackable than the upper left edges of the tank, the aimpoint will slide toward the back of the tank (indicated by the cross 68). As the aimpoint slides, a higher percentage of clutter will be incorporated into future correlation reference snapshots, to the exclusion of subimages of the target. Eventually, the aimpoint may be off the target altogether. The rate of sliding is a function of target velocity, the amount of clutter in any one tracker subimage, and the infrared ("IR") strength of the clutter relative to the target if the tracker is IR sensitive. The same scenario will occur if, instead of using four correlator groups as depicted, any number of arbitrary subimages were employed.
The reason why the aimpoint can slide in scenarios like FIG. 8 is that the aimpoint is computed by assuming that all the individual subimages have the same magnification around the aimpoint. If the magnification of one or more subimages are bad, then the magnification errors will bias the aimpoint. If the bad subimage tracker can be detected and removed, the aimpoint may be recomputed based on the positions of the remaining good subimages.
Bad subimages may be detected by looking for large differences between the measured subimage location x and the estimated measurement x:
Each estimated subimage location is determined from the estimated aimpoint. Typically this is accomplished by adding the product of an estimated magnification factor and the normalized offset vector to the estimated location of the aimpoint. The estimated aimpoint is itself determined by the location of all subimages, both good and bad, as described above.
FIG. 9 depicts the location of the measurements and the estimated measurements in a scenario such as FIG. 8. The measured locations of the subimages, x.sub.1, x.sub.2, x.sub.3, and x.sub.4 are indicated by squared points while the estimated measurements, x.sub.1, x.sub.2, x.sub.3, and x.sub.4 are indicated by circled points. The estimated and true aimpoints are indicated by x.sub.A and x.sub.A respectively. Here, the residual, or error, .epsilon..sub.4 of the upper left subimage x.sub.4 -x.sub.4 is larger and of a different direction compared to the other three subimages. A robust statistical procedure may be used to detect the drift where:
MAD stands for the median absolute deviation. Other mathematical differences may also be used to test the behavior of each subimage.
The MAD is a robust estimator of the scale or spread of a distribution of values. The ratio MAD/0.6745 approaches the standard deviation for a Gaussian distribution as N becomes large. A robust T-test may be used of the form: ##EQU4## If T.sub.i is greater than a predetermined displacement threshold, then the tracker measurement is considered bad and is deleted from the vector of subimage locations x. The T.sub.i statistic will not be corrupted by bad subimages as long as the majority of the trackers are good. The test is performed for both the x and y image dimensions.
The robust T-test of Eq. (13) would eventually detect and reject the bad subimage of FIGS. 8 and 9. By recomputing the aimpoint and magnification estimates using a subset of the subimages of only the good measurements, the re-estimated aimpoint will be much closer to the true aimpoint.
FIG. 10 depicts the steps described above and is a detailed diagram of step 58 of FIG. 7. An aimpoint is estimated in block 70 initially using all subimages available to the tracker at a certain time. In block 72, each subimage location is estimated using the aimpoint estimate of block 70. This calculation may typically be accomplished by adding the product of a range normalized distance and magnification factor or factors (TIMF) to the aimpoint location. Each subimage has a range normalized distance that relates its location to the location of the aimpoint. A difference is calculated in block 74 between the estimated subimage location from block 72 and the measured location of each subimage in block 56 (FIG. 7). This difference is then statistically compared to a threshold value in block 76. If the difference exceeds the threshold value, then the subimage is rejected and deleted from the memory or database containing all of the subimages. In block 78, a new aimpoint is calculated using the revised aimpoint database.
ONE-DIMENSIONAL SUBIMAGE TRACKING
FIG. 11 depicts a target having a subimage associated with a line. Here, the target is a tall building, its aimpoint is indicated by a cross-hairs and it is tracked by two subimages indicated by small squares. The first subimage is associated with the right edge of the building. The second subimage is associated with a corner on the left side of the building. In practice, the aimpoint is determined by more than two subimages. The first subimage may be mistakenly found by the tracker at a subsequent time anywhere along the left edge of the building. The second subimage, however, will remain at the corner of the building. The first and second subimages are consequently called "one-dimensional" and "two-dimensional" subimages, respectively. The aimpoint will move from its original position if the one-dimensional subimage shifts up or down the left edge and nothing else is adjusted. The aimpoint may be held constant if the one-dimensional subimage is identified and if the tracker procedures are accordingly modified.
Identification of One-Dimensional Subimages
FIGS. 12a and 12b graphically depict a correlator test associated with a one-dimensional and a two-dimensional subimage, respectively. The correlator test used in the preferred embodiment is a modification of the correlator test used to locate subimages in the field of view of the tracker after an initial time. In the original correlator test, the absolute value of the difference between the gray level data of a reference image and of a subsequently acquired image is determined. The gray level data is generally the luminosity of the subimage at a particular pixel location. The luminosity is the infrared "brightness" of the image in the preferred embodiment. Other wavelengths of electromagnetic radiation may be used to track a target and, hence, may be compared according to the disclosed invention. The difference is determined for each pair of pixels in the two images within an image area. The first image area is centered around the reference subimage and the second image area is initially centered around the predicted location of the subimage at the subsequent time. The summed difference is calculated several times. Each time, the center of the subimage in the subsequently acquired image is indexed or moved a predetermined amount. Each pixel is thereby shifted a predetermined amount, typically one pixel. This causes a different set of gray level data pairs to be subtracted from one another. The set having the smallest summed difference indicates the actual location of the subimage in the subsequently acquired image. The center of the actual subimage is then used as the location of the subimage at the later time by the tracker for mid-course tracking.
The correlator test may be modified to determine the dimensionality of a particular subimage (1-D versus 2-D). In this modified test, the gray level data of the subimage is compared to the indexed gray level data of the subimage at the same time. The previously described difference for each pixel pair is calculated, summed over the image and saved. In FIG. 12a, a subimage having a circular shape is indexed through an image space and compared to the original subimage. As depicted, the subimage is indexed through image space having roughly a diamond shape. The diamond is seven pixels high by seven pixels wide and is further described in connection with FIG. 14. This results in twenty-eight sets of comparisons. (The comparison when the subimage is indexed to its original position is always zero.) In FIG. 12b, a part of an edge is depicted. It is also indexed through an image space having a diamond shape. Only the four most extreme locations of each indexed subimage is depicted in each FIGURE. The indexed subimages in FIGS. 12a and 12b are depicted as dashed circles or edges. The actual subimages are depicted as a solid circle or edge.
FIGS. 13a and 13b graphically depict the surfaces generated by the correlator tests of FIGS. 12a and 12b, respectively. Each surface depicts the summed differences for each indexed location (x, y) as a value along the z-axis. A two-dimensional subimage like the one depicted in FIG. 12a generates the bowl shaped surface depicted in FIG. 13a. A one-dimensional subimage like the one depicted in FIG. 12b generates the V-shaped surface depicted in FIG. 13b.
The shape of the surface generated by the twenty-eight correlator differences may be used to determine the dimensionality of the subimage. The surface is statistically fit to a model paraboloid having the form:
where C(i,j) is the summed difference corresponding to the shifted subimage centered at (i,j) and a, b, c and d are constants. The indices i and j are analogous to the x and y coordinate axes, respectively, depicted in FIGS. 12a through 13b. The constants may be determined by a linear least squares curve fit: ##EQU5## where C is a 28.times.1 data vector containing the values of C(i,j) and G is a 3.times.28 matrix of predetermined constants. The ordering of C and the extent of subimage indexing undertaken in FIGS. 12a and 12b are depicted in FIG. 14. The contents of G are depicted in FIG. 15. Once the constants a, b and c are determined, the eigenvalues of the 2.times.2 matrix: ##EQU6## are computed. The maximum and minimum eigenvalues, .lambda..sub.max and .lambda..sub.min, are calculated according to the relations: ##EQU7## The ratio of these two eigenvalues, .lambda..sub.max /.lambda..sub.min, may be used to distinguish one-dimensional subimages from two-dimensional subimages. If the ratio is greater than a predetermined constant such as 12, then the subimage is considered a one-dimensional subimage.
TRACKING WITH ONE-DIMENSIONAL SUBIMAGES
FIG. 16 mathematically depicts a subimage i at (x.sub.i, y.sub.i) associated with a line with a first (x and y) and a second (x' and y') coordinate system. Both coordinate systems are centered on the aimpoint at (x.sub.a, y.sub.a). The line is indicated by hatching on one of its sides.
If Q is a linear transformation representing an orthonormal basis set, then
represents the transformation of vector x into x' whose components are defined with respect to the x' and y' coordinate frame. The matrix Q.sup.T comprises the maximum and minimum eigenvectors, Q.sub.max.sup.T and Q.sub.min.sup.T, of the 2.times.2 matrix: ##EQU8## Additionally, Q.sup.-1 equals Q.sup.T. Therefore,
If .delta.x.sub.i =x.sub.i -x.sub.a then the subimage may be related to the aimpoint in the eigenspace represented by Q through the relations:
This may be rewritten as
where d.sub.i,max and n.sub.i,max are the components of the normalized distance between the aimpoint and the ith subimage and of the noise vector, respectively, which are along or parallel to the maximum eigenvector. This may be further rewritten as: ##EQU10## where ##EQU11## For each one-dimensional subimage i, the vector [q.sub.i,1,max q.sub.i,2,max d.sub.i,max ] may be directly substituted into the ith row of the H matrix described above (Eq. (6)). In particular, q.sub.i,1,max is substituted into the first column, ith row, q.sub.i,2,max is substituted into the second column, ith row, and d.sub.i,max is substituted into the third column, ith row. For a condition where there are four subimages, the second and third subimages being one-dimensional and the remaining subimages being two-dimensional, the H matrix would be ##EQU12## For every one-dimensional subimage i, the 2ith row is removed without substitution. Alternatively, for such one-dimensional subimage i, the vector [q.sub.i,1,max q.sub.i,2,max d.sub.i,max ] may be substituted into the 2ith row and the ith row removed. In this manner, a composite H matrix may be assembled by which two-dimensional subimages may be simultaneously tracked with one-dimensional subimages. The values for x.sub.i,max and d.sub.i,max are calculated by multiplying x.sub.i and d.sub.i, respectively, by the maximum eigenvector q.sub.i,max.sup.T. Given this specification, the calculation of the eigenvectors will be apparent to one skilled in the art. The H matrix is then used to solve for the vector .theta. as described above (see Eq. (9)).
Error Checking with One-Dimensional Subimages
FIG. 17 further depicts the subimage depicted in FIG. 16. The residual .epsilon..sup.T for a 1-D tracker is a transformed version of the residual in the original coordinate system.
The quantity q.sub.i,max.sup.T (x.sub.i -x.sub.i) is graphically represented in FIG. 17 as dx'. Bad subimages may therefore be tested and rejected as described for the general case after being multiplied by the maximum eigenvector. For 1-D subimage rejection, the quantity .epsilon..sub.i.sup.T is substituted for .delta.x.sub.i in equations (11), (12) and (13), and the same T-test is performed.
FIG. 18 depicts part of the flow chart of FIG. 7. Block 52 comprises blocks 80, 82 and 84. In these blocks, the tracker acquires all subimages, categorizes each subimage as either 1-D or 2-D and for each subimage, loads the H matrix with either a set of 2-D values (see Eq. 6) or a set of 1-D values (see, e.g., Eq. 31), as appropriate. Each of these steps is more fully described above.
Appendices A-H contain FORTRAN computer code disclosing one embodiment of the disclosed invention. In particular, Appendix A discloses the subroutine TRTEST and related functions for determining the dimensionality of a subimage. Appendix B discloses the subroutine DBADD for adding a subimage to the tracker's database. Appendix C discloses the subroutine EIGENV for calculating the eigenvalues and eigenvectors of a selected subimage. Appendix D discloses subroutine TRKNORNG2D for computing the aimpoint location using a single magnification model. Appendix E discloses the subroutine INTJITTER for rejecting bad subimages associated with an aimpoint under the subroutine TRKNORNG2D. Appendix F discloses the subroutine LEASTSQ for calculating the least-squares estimate of a parameter vector. Appendix G discloses the common block variable declaration PROCPARM for the previous subroutines. Appendix H discloses a library of subroutines useful primarily for matrix math called by the previous subroutines.
Although the present invention and its advantages have been described in detail, it should be understood that various changes, substitutions and alterations can be made herein without departing from the spirit and scope of the invention as defined by the appended claims. ##SPC1##
Claims
What is claimed is:
1. A method for tracking an aimpoint, comprising the steps of: acquiring an aimpoint on a target and a set of subimages arbitrarily associated with the aimpoint by a sensor in a field of view; for a first time, for each subimage, determining whether the subimage is one-dimensional or two-dimensional; for the first time, calculating a normalized distance from each subimage to the aimpoint; for the first time, for each one-dimensional subimage, determining a maximum eigenvector for the subimage; for the first time, for each one-dimensional subimage, calculating a component of the normalized distance associated therewith which is along the maximum eigenvector therefor; for a second time, reacquiring at least one subimage; and estimating a subsequent location of the aimpoint based on the subsequent location of the subimage and on the normalized distance therefor and, if said reacquired subimage is one-dimensional, further based upon the maximum eigenvector and normalized distance component therefor.
2. The method of claim 1, wherein said estimating step further comprises the steps of calculating the subsequent location of the aimpoint using a least squares technique on a vector for all reacquired subimages, said vector including normalized distances for each reacquired two-dimensional subimage and a maximum eigenvector and normalized distance component for each reacquired one-dimensional subimage.
3. The method of claim 1, wherein said step of determining whether the subimage is one-dimensional or two-dimensional further comprises the steps of: generating a set of data for each subimage, each set of the data representative of the difference between the gray level data of each subimage and the gray level data of the same subimage indexed through an image space adjacent the subimage; fitting the set of data to a paraboloid; determining maximum and minimum eigenvalues from the constants of the parabola; and calculating the ratio of the maximum eigenvalue to the minimum eigenvalue.
4. A method for tracking an aimpoint on a target comprising the steps of: selecting an aimpoint on the target; for a first time, acquiring a set of subimages of the target arbitrarily associated with the aimpoint using predetermined subimage trackability criteria, using an image sensor operable to change its displacement relative to the target; for each subimage, determining whether the subimage is one-dimensional or two-dimensional, including the substeps of generating a set of data for each subimage, each set of the data representative of the difference between the gray level data of each subimage and the gray level data of the same subimage indexed through an image space around the subimage; fitting the set of data to a paraboloid; determining maximum and minimum eigenvalues from the constants of the parabola; calculating the ratio of the maximum eigenvalue to the minimum eigenvalue; calculating the geometric relationship of each subimage to the aimpoint, taking into account whether the subimage is one-dimensional or two-dimensional; between a first time and a second time, changing the displacement of the image sensor relative to the target; reacquiring at least one of the subimages at a second time using the sensor; and calculating a location of the aimpoint at said second time using the geometric relationship between the at least one subimage and the aimpoint.
5. The method of claim 4, wherein said reacquiring step further comprises the step of reacquiring a set of subimages with a sensor mounted on a missile.
6. The method of claim 4, wherein said second calculating step further comprises the step of estimating the second aimpoint based on a least squares technique applied to a vector including normalized distances for each reacquired two-dimensional subimage and a normalized distance component along a maximum eigenvector for each reacquired one-dimensional subimage.
7. A method for rejecting subimages erroneously associated with an aimpoint, comprising the steps of: determining the locations of a set of subimages associated with an aimpoint at a first time; for each subimage, determining whether the subimage is one-dimensional or two-dimensional; estimating the location of the aimpoint at a second time based on the locations of the set of subimages at the second time and on a geometric relationship between the subimages and the aimpoint at the first time; estimating a set of locations of the subimages which should have occurred at the second time, each estimated location based on the estimated aimpoint at the second time; setting a displacement threshold; for each one-dimensional subimage, calculating a residual based on a maximum eigenvector and a difference between an estimated location of the one-dimensional subimage and a corresponding measured location therefor; for each two-dimensional subimage calculating a difference between each estimated two-dimensional subimage location and the corresponding measured location therefor; and rejecting each subimage where a predetermined function of the residual or difference thereof exceeds the displacement threshold.
8. The method of claim 7, and further comprising the step of recalculating the location of the aimpoint based on all of the remaining nonrejected subimages of the set.
9. The method of claim 7, wherein the recalculated location of the aimpoint is calculated using a least squares technique.
10. The method of claim 7, wherein the predetermined function is a T-test.
11. The method of claim 7, wherein said step of determining whether the subimage is one-dimensional or two-dimensional further comprises the steps of: generating a set of data for each subimage, each set of the data representative of the difference between the gray level data of each subimage and the gray level data of the same subimage indexed through an image space around the subimage; fitting the set of data to a paraboloid; determining maximum and minimum eigenvalues from the constants of the paraboloid; and calculating the ratio of the maximum eigenvalue to the minimum eigenvalue.
12. A method for tracking an aimpoint, comprising the steps of: acquiring an aimpoint and a set of subimages arbitrarily associated with the aimpoint at a first time; for each subimage, determining whether the subimage is one-dimensional or two-dimensional; calculating the normalized distance from each two-dimensional subimage to the aimpoint; for each one-dimensional subimage, determining a maximum eigenvector for the subimage; for each one-dimensional subimage, calculating a component of the normalized distance from the aimpoint to the one-dimensional subimage which is parallel to the maximum eigenvector therefor; at a later time reacquiring a plurality of subimages; estimating a subsequent location of the aimpoint based on the reacquired subimage and on the normalized distances, and, if a reacquired subimage is one-dimensional, based on the maximum eigenvector therefor; setting a displacement threshold; estimating a set of locations of the subimages based on the subsequent location of the aimpoint and on the normalized distance and, to the extent that any of the subimages are one-dimensional, based on the maximum eigenvectors therefor; calculating the difference between the estimated location and the reacquired location of each subimage; rejecting each reacquired subimage where a predetermined function of the difference exceeds the displacement threshold; and re-estimating the aimpoint based on the remaining reacquired subimages.
13. The method of claim 12, wherein said step of determining whether each subimage is one-dimensional or two-dimensional further comprises the steps of: generating a set of data for each subimage, each set of the data representative of the difference between the gray level data of each subimage and the gray level data of the same subimage indexed through an image space surrounding the subimage; fitting the set of data to a paraboloid; determining maximum and minimum eigenvalues from the constants of the parabola; and calculating the ratio of the maximum eigenvalue to the minimum eigenvalue.
14. A method for tracking an aimpoint, comprising the steps of: acquiring an aimpoint on a target, and a set of subimages arbitrarily associated with the aimpoint, by a sensor in a field of view; for a first time and for each subimage, determining whether the subimage is one-dimensional or two-dimensional by performing the following substeps: for each of a plurality of pixels of the subimage, acquiring a gray level datum of pixel brightness; shifting the subimage by a predetermined number of pixels; for each of a plurality of pixels in the shifted subimage, acquiring a gray level datum of pixel brightness; correlating each pixel in the subimage with a corresponding pixel in the shifted subimage to create a gray level data pair; for each gray level data pair, subtracting the gray level of one pixel from the gray level of the other pixel in the pair to obtain a gray level difference; summing the gray level differences for all of the gray level data pairs between the subimage and the shifted subimage to obtain a sum of differences therefor; storing the summed differences obtained for the subimage and the shifted subimage; repeating said substeps of shifting, acquiring a gray level datum for each pixel in the shifted subimage, correlating, obtaining a gray level difference and obtaining a sum of differences, and storing for each of a plurality of shifted subimages, each as compared with the subimage, over a predetermined image space surrounding the subimage; using the stored sums of differences, finding the constants a, b and c in the following relation: where C(i,j) is the sum of differences corresponding to a shifted subimage centered at location (i,j) and d is a constant; from constants a, b and c, determining maximum and minimum eigenvalues for C(i,j); determining a ratio of the maximum eigenvalue to the minimum eigenvalue; finding that the subimage is one-dimensional if the ratio exceeds a predetermined constant; finding that the subimage is two-dimensional if the ratio does not exceed a predetermined constant; for the first time, for each two-dimensional subimage, determining a normalized distance from the two-dimensional subimage to the aimpoint; for the first time, for each one-dimensional subimage, determining a maximum eigenvector from the found constants a, b and c; for the first time, for each one-dimensional subimage, determining a component of the normalized distance from the one-dimensional subimage to the aimpoint along the maximum eigenvector therefor; for a second time, reacquiring at least one of the subimages; and estimating a location of the aimpoint at the second time based on the location of said at least one subimage at the second time, the normalized distance of said at least one subimage to the aimpoint, and, if said at least one subimage is one-dimensional, the maximum eigenvector therefor and said component of the normalized distance along the last said maximum eigenvector.
15. The method of claim 14, and further comprising the step of reacquiring a plurality of subimages for the second time, the location of the aimpoint estimated using a plurality of subimages.
16. A guidance system for tracking an aimpoint comprising: a sensor for initially acquiring an aimpoint and for periodically acquiring a set of subimages arbitrarily associated with the aimpoint; a processor for calculating the normalized distance from the first set of subimages to the aimpoint, for determining the dimensionality of each subimage and for estimating subsequent locations of the aimpoint based on the periodically acquired subimages, on the normalized distances and, where a subimage is one-dimensional, on a maximum eigenvector of the subimage; and memory for storing the normalized distances and eigenvectors.
17. The guidance system of claim 16, further comprising means for moving the sensor towards each of the subsequent locations of the aimpoint.
18. A missile, comprising: a sensor for initially acquiring an aimpoint and for periodically acquiring a set of subimages arbitrarily associated with the aimpoint; a processor for calculating the normalized distance from the first set of subimages to the aimpoint, for determining the dimensionality of each subimage and for estimating subsequent locations of the aimpoint based on the periodically acquired subimages, on the normalized distances and, where a subimage is one-dimensional, on the maximum eigenvector associated with the subimage; memory for storing the normalized distances and eigenvectors; fins for guiding the missile responsive to the estimated aimpoint locations; and a motor for propelling the missile.
Patent Citations (6)
Non-Patent Literature (5)
- Collection of presentation materials prepared by the Applicants on Jun. 26, 1991 for presentation to the U.S. Army technical staff.
- A computer printout of the results of a patent search conducted Aug. 20, 1991 by Texas Instruments' library personnel.
- Blackman, Samuel, Multiple-Target Tracking with Radar Applications, Artech House, Inc., 1986, pp. 309-328.
- Huber, Peter, Robust Statistics, John Wiley & Sons, Inc., 1981, pp. 107-108.
- Liu, Zhili, "New Image Tracking Algorithm for Fuzzy-Relaxation Matching of Point Patterns", Hongwai Yanjiu, vol. 8, No. 5, 1978, pp. 349-354 (translated into English by the Foreign Technology Division of defense Technical Information Center, Defense Logistics Agency).