Correlation tracker breaklock detection

Abstract

A system and method for detecting breaklock for correlation trackers. The inventive system includes a first circuit (50) for computing energy EL in a residual image, a second circuit (52) for computing energy EI in an input image, and a third circuit (54) for computing a residual metric RL based on the residual image energy EL scaled by the input image energy EI. In the illustrative embodiment, the system further includes a fourth circuit (56) for comparing the residual metric RL to a threshold TR, and a fifth circuit (58) for setting a breaklock signal based on the output of the fourth circuit (56).

Description

BACKGROUND OF THE INVENTION

[0002] 1. Field of the Invention

[0003] The present invention relates to imaging systems. More specifically, the present invention relates to correlation based imaging target trackers.

[0004] 2. Description of the Related Art

[0005] An autotracker is a device which locates a target in each of a sequence of images and generates commands to maintain a sensor line-of-sight on the target. Correlation trackers generate servo commands based on the position of the target. The tracker measures this position by finding the shift of the input image that maximizes its cross-correlation with a reference image formed by averaging recent past input images.

[0006] During a target tracking operation, it often happens that the line-of-sight from the sensor to the target becomes temporarily obscured. For example, in an air-to-ground scenario, a cloud or a building may pass between the sensor and the target temporarily blocking the line-of-sight. One important function of the autotracker is to determine when the target has been obscured, thereby detecting a ‘breaklock’ condition. The tracker is then commanded into a ‘coast mode’ until the target is again visible. In coast mode the sensor is pointed open-loop at the predicted position of the target based on its recent motion. While in coast mode the tracker must continually monitor the input image to determine when the target is again visible so that closed-loop tracking may resume.

[0007] Prior attempts to detect breaklock made use of either: 1) a threshold on a statistical average of the residuals after cross-correlation (see, for example, “Developing a Real-Time, Robust, Video Tracker,” K. Plakas, E. Trucco (0-7803-6551-8/00 IEEE) or 2) a threshold on the change in the difference between the input image and the correlation reference map. Unfortunately, these approaches have proven to be too sensitive to certain image disturbances which commonly occur in dynamic tracking situations. This is due to the reliance by these approaches on an assumption that the difference between the input image and reference map pixel values has a Gaussian density. While this may be true in an idealized situation, in practical tracking situations there are several types of image disturbances which cause the Gaussian assumption to falter. One such disturbance is due to a change of the sensor position relative to the tracked target which causes the target's image to change in size and shape. Another is due to uncommanded sensor line-of-sight deviations such as jitter which cause image blurring. These image disturbances will cause the existing breaklock detection methods to falsely declare a breaklock condition when continued tracking is still feasible.

[0008] Hence, a need exists in the art for an improved system or method for detecting breaklock for correlation trackers.

SUMMARY OF THE INVENTION

[0009] The need in the art is addressed by the system and method for detecting breaklock for correlation trackers of the present invention. The inventive system includes a first circuit for computing energy EL in a residual image, a second circuit for computing energy EI in an input image, and a third circuit for computing a residual metric RL based on the residual image energy EL scaled by the input image energy EI.

[0010] In the illustrative embodiment, the system further includes a fourth circuit for comparing the residual metric RL to a threshold TR, and a fifth circuit for setting a breaklock signal based on the output of the fourth circuit.

[0011] The present invention addresses the problems associated with the prior art by comparing the residual energy to the energy in the input image, rather than the current method of comparing the residual energy to an absolute threshold. By scaling the residual energy by the input image energy, a more robust statistic is obtained which allows closed-loop tracking to continue as long as even a small amount of match exists between the input and reference images. The invention has been designed to quickly and accurately determine actual breaklock conditions such as obscurations and extreme uncommanded line-of-sight deviations while enabling continued tracking through high sensor-to-target relative position rates and sensor line-of-sight jitter.

BRIEF DESCRIPTION OF THE DRAWINGS

[0012] FIG. 1 is a block diagram of an autotracker system.

[0013] FIG. 2 is a block diagram of the components of an autotracker designed in accordance with the teachings of the present invention.

[0014] FIG. 3 is a flow chart illustrating the breaklock detection method of the present invention.

[0015] FIG. 4 is a block diagram showing a hardware implementation of the autotracker system of the present invention.

DESCRIPTION OF THE INVENTION

[0016] Illustrative embodiments and exemplary applications will now be described with reference to the accompanying drawings to disclose the advantageous teachings of the present invention.

[0017] While the present invention is described herein with reference to illustrative embodiments for particular applications, it should be understood that the invention is not limited thereto. Those having ordinary skill in the art and access to the teachings provided herein will recognize additional modifications, applications, and embodiments within the scope thereof and additional fields in which the present invention would be of significant utility.

[0018] FIG. 1 is a block diagram of an autotracker system 10. A monochromatic sensor 12, mounted in a gimbal 14 generates a two dimensional image, I(r,c), of the target. Example sensors are an infrared (IR) sensor or a TV camera. This image is processed by the autotracker 16, which generates commands to the servo motors 18 to maintain the sensor line-of-sight on the target. In a correlation tracker, the position of the target is measured by finding the shift of the input image that maximizes its cross-correlation with a reference image formed by averaging recent past input images.

[0019] The present invention is a system and method for breaklock detection for use in correlation autotracker systems. It uses statistics generated by the correlation process to compute a residual metric that is used to determine, while tracking, when the target is no longer present in the field of view, and, while coasting, when the target is again visible in the field of view. The invention addresses the problems of the prior art by comparing the residual energy to the energy in the input image, rather than the current method of comparing the residual energy to an absolute threshold. By scaling the residual energy by the input image energy, a more robust statistic is obtained which allows closed-loop tracking to continue as long as even a small amount of match exists between the input and reference images. The invention has been designed to quickly and accurately determine actual breaklock conditions such as obscurations and extreme uncommanded line-of-sight deviations while enabling continued tracking through high sensor-to-target relative position rates and sensor line-of-sight jitter.

[0020] FIG. 2 is a block diagram of the components of an autotracker 16 designed in accordance with the teachings of the present invention. A correlation pixel processing function 20 maintains a reference map and cross-correlates this map with each input image to determine the track errors and a residual metric. The track errors are used by the closed-loop control function 22 to compute servo commands to keep the tracked target in the center of the sensor field of view. The residual metric, the subject of this patent, is used by status processing 24 to derive a breaklock signal. This breaklock signal is used by the closed-loop control function 22 to determine when it should point the sensor based on the track errors (closed-loop mode) or when it should ignore the track errors and point using the predicted target position (coast mode).

[0021] While this method can be applied to any correlation tracker, the invention will now be described with reference to an illustrative application to the Fitts correlation pixel processing algorithm described in U.S. Pat. No. 4,133,004, issued Jan. 2, 1979, entitled “Video Correlation Tracker”, the teachings of which are incorporated herein by reference. In accordance with the Fitts teachings, a correlation pixel processing algorithm generates a reference map, denoted M(r,c), in pixel format by averaging input images prior to the current time t. (It is assumed herein that all images mentioned have Nrows and Ncols columns of pixels for a total of Npixels=Nrows*Ncols pixels. Thus, the row numbers r, have values in the range from 1 to Nrows inclusive, and the column numbers, c, have values in the range from 1 to Ncols inclusive.)

[0022] The correlation pixel processing 20 correlates the reference map M(r,c) with the input image at time t, denoted I(r,c), to determine the track error, denoted 1$\delta =\left[\begin{array}{c}{\delta }_r\\ {\delta }_c\end{array}\right],$

[0023] which is the estimated shift of the input image relative to the reference image. As described in the Fitts correlation patent, in the process of computing the track error the Fitts algorithm computes gradient images that are approximations to the image spatial derivatives in the row and column directions. The gradient image in the row direction, denoted Gr(r,c), is given by: 2$\begin{array}{cc}{G}_r\left(r,c\right)=\frac{M\left(r+1,c\right)-M\left(r-1,c\right)}{2}& \left[1\right]\end{array}$

[0024] and the gradient image in the column direction, denoted Gc(r,c) is given by: 3$\begin{array}{cc}{G}_c\left(r,c\right)=\frac{M\left(r,c+1\right)-M\left(r,c-1\right)}{2}& \left[2\right]\end{array}$

[0025] The residual image, denoted L(r,c), is defined to be the difference between the input image and the reference map shifted by δ′, where 4${\delta }^{\prime }=\left[\begin{array}{c}{\delta }_r^{\prime }\\ {\delta }_c^{\prime }\end{array}\right]$

[0026] is the computed track error before drift compensation, as described in the Fitts algorithm patent. The shifted reference map, denoted M′(r,c), may be approximated using a first order Taylor expansion by:

M ′(r,c)=M(r,c)+Gr(r,c)δ′r+Gc(r,c)δ′c  [3]

[0027] Thus, the residual image can be approximated by:

L (r,c)=I(r,c)−M′(r,c)=I(r,c)−M(r,c)−Gr(r,c)δ′r−Gc(r,c)δ′c,  [4]

or

L (r,c)=D(r,c)−Gr(r,c)δ′r−Gc(r,c)δ′c  [5]

[0028] where D(r,c) is the difference image defined by:

D (r,c)=I(r,c)−M(r,c)  [6]

[0029] The energy in the residual image, denoted EL, gives an indication of the relative amount of unexplained variation in the residual image, and is defined as: 5$\begin{array}{cc}{E}_L=\sum _{\left(r,c\right)\in \mathrm{Gate}}{\left(L\left(r,c\right)-\stackrel{\_}{L}\right)}^2,& \left[7\right]\end{array}$

[0030] where {overscore (L)} is the average of the residual image pixel values given by: 6$\begin{array}{cc}\stackrel{\_}{L}=\frac{1}{N_{\mathrm{pixels}}}·\sum _{\left(r,c\right)\in \mathrm{Gate}}L\left(r,c\right)& \left[8\right]\end{array}$

[0031] and Gate represents the set of (r,c) values for which 1≦r≦Nrows and 1≦c≦Ncols.

[0032] Similarly, the energy in the input image, denoted EI, is defined to be: 7$\begin{array}{cc}{E}_I=\sum _{\left(r,c\right)\in \mathrm{Gate}}{\left(I\left(r,c\right)-\stackrel{\_}{I}\right)}^2,& \left[9\right]\end{array}$

[0033] where {overscore (I)} is the average of the input image pixel values and is given by: 8$\begin{array}{cc}\stackrel{\_}{I}=\frac{1}{N_{\mathrm{pixels}}}·\sum _{\left(r,c\right)\in \mathrm{Gate}}I\left(r,c\right)& \left[10\right]\end{array}$

[0034] The residual metric, denoted RL, is then defined to be: 9$\begin{array}{cc}{R}_L=\frac{E_L}{E_I}& \left[11\right]\end{array}$

[0035] and represents the relative amount of the energy in the input image that is not explained by modeling it as the shifted reference image.

[0036] In practice the energy in the residual image can be approximated by the energy in the difference image because the image shifts are relatively small, and the change they induce can often be neglected. Let the energy in the difference image, denoted ED, be defined as: 10$\begin{array}{cc}{E}_D=\sum _{\left(r,c\right)\in \mathrm{Gate}}{\left(D\left(r,c\right)-\stackrel{\_}{D}\right)}^2& \left[12\right]\end{array}$

[0037] where {overscore (D)} is the average of the difference image pixel values given by: 11$\begin{array}{cc}\stackrel{\_}{D}=\frac{1}{N_{\mathrm{pixels}}}·\sum _{\left(r,c\right)\in \mathrm{Gate}}D\left(r,c\right)& \left[13\right]\end{array}$

[0038] The difference residual metric, denoted RD, is then defined to be: 12$\begin{array}{cc}{R}_D=\frac{E_D}{E_I}& \left[14\right]\end{array}$

[0039] The residual metric, either RD or RL, is used by the status processing function 24 to determine the value of the breaklock signal. If the shifted reference image exactly matches the input image then the residual metric will very nearly equal zero. As the degree of match decreases the residual metric grows in size. Values of this metric close to or greater than one indicate a complete mismatch between the reference image and the input image. Therefore, a threshold, TR, on the residual metric is used to derive the breaklock signal. If the residual metric is less than or equal to TR, the match between the reference image and the input image is considered good enough to continue tracking and the breaklock signal is set to FALSE indicating that closed-loop tracking is to continue. If the residual metric is greater than TR, then the match has significantly degraded and the breaklock signal is set to TRUE indicating that the closed-loop control should enter coast mode. Empirical experience has shown that tracking can reliably continue with TR values up to 0.9.

[0040] FIG. 3 is a flow chart illustrating the breaklock detection method 60 of the present invention. First, the energy EL in the residual image is determined (50), and the energy EI in the input image is determined (52). In the illustrative embodiment, EL is computed using Eqn. 7, or estimated using Eqn. 12. In the illustrative embodiment, EI is computed using Eqn. 9. Next, the residual metric RL=EL/EI is computed (54). Then, the residual metric RL is compared to a threshold TR (56). Finally, if the residual metric RL was greater than the threshold TR, then the Breaklock Signal is set to TRUE; otherwise, it is set to FALSE (58).

[0041] FIG. 4 is a block diagram showing a hardware implementation of the autotracker system 10 of the present invention. A sensor 12 mounted in a gimbal 14 receives visible or infrared images which are then processed by a video processor 30. The video processor 30 formats the received images into digital video and, for infrared, performs non-uniformity correction and dead cell replacement. The digital images are then sent to an image processor 32 which computes the track errors and the residual metric. The track errors are used by a servo control processor 34 to generate the torque commands for the sensor gimbal hardware 14 to keep the tracked target in the center of the sensor field of view. The residual metric is used by a system control processor 36 to derive a breaklock signal. This breaklock signal is used by the servo control processor 34 to determine when it should point the sensor based on the track errors (closed-loop mode) or when it should ignore the track errors and point using the predicted target position (coast mode). The system control processor 36 controls the operation of the entire autotracker system and may interface to an aircraft.

[0042] Thus, the present invention has been described herein with reference to a particular embodiment for a particular application. Those having ordinary skill in the art and access to the present teachings will recognize additional modifications, applications and embodiments within the scope thereof.

[0043] It is therefore intended by the appended claims to cover any and all such applications, modifications and embodiments within the scope of the present invention.

[0044] Accordingly,

Claims

1. A system for detecting breaklock for correlation trackers comprising: first means for computing energy EL in a residual image; second means for computing energy EI in an input image; and third means for computing a residual metric RL based on the residual image energy EL scaled by the input image energy EI.

2. The invention of claim 1 wherein said system further includes fourth means for comparing said residual metric RL to a threshold TR.

3. The invention of claim 2 wherein said system further includes fifth means for setting a breaklock signal based on the output of said fourth means.

4. The invention of claim 2 wherein said breaklock signal is set to TRUE if the residual metric RL is greater than the threshold TR and FALSE otherwise.

5. The invention of claim 1 wherein said residual image energy EL is computed using the equation

6. The invention of claim 1 wherein said residual image energy EL is approximated by the energy ED in a difference image D(r,c).

7. The invention of claim 6 wherein said difference image energy ED is computed using the equation

8. The invention of claim 1 wherein said input image energy EI is computed using the equation

9. The invention of claim 1 wherein said residual metric RL is computed using the equation RL=EL/EI.

10. An autotracking system comprising: a sensor mounted on a gimbal for receiving an input image; an image processor which computes track errors, and a residual metric RL based on a residual image energy EL scaled by an input image energy EI; a servo control processor which computes commands for the sensor gimbal based on the track errors; and a system control processor which generates a breaklock signal for the servo control processor based on the residual metric RL.

11. A method for detecting breaklock for correlation trackers including the steps of: computing the energy EL in a residual image; computing the energy EI in an input image; computing a residual metric RL based on the residual image energy EL scaled by the input image energy EI; comparing said residual metric RL to a threshold TR; and setting a breaklock signal based on said comparison.