Efficient track-before detect algorithm with minimal prior knowledge

1. A system for tracking a target in images, the system comprising:

a processor;

memory, operatively coupled to the processor, wherein the memory comprises computer-executable instructions;

an output module, operatively coupled to the processor; and

an input module, operatively coupled to the processor, for receiving a stream of change-detected images, wherein each of the change-detected images comprises a plurality of pixels, wherein each of the change-detected images has been processed to remove static objects therefrom;

wherein the computer-executable instructions, when executed by the processor, cause the processor to perform a plurality of acts, comprising:

processing the stream of change-detected images via a plurality of different recursive motion model kernels in parallel to provide a plurality of kernel outputs for a pixel of a change-detected image in the change-detected images, wherein each recursive motion model kernel has a respective motion model, and further wherein each recursive motion model kernel is configured to generate an output in the outputs that is indicative of whether the pixel represents a target that is moving in conformance with the respective motion model of the recursive motion model kernel, the output comprises a velocity estimate for the pixel and an energy estimate for the pixel; and

fusing the plurality of kernel outputs to produce a fused estimate, wherein the fused estimate comprises a fused velocity estimate and a fused energy estimate that is indicative of motion of the target in the stream of change-detected images.

2. The system of claim 1 , wherein the energy estimate is indicative of direction of movement of the target.

3. The system of claim 2 , wherein the fused estimate comprises a highest estimated energy from among energy estimates in the kernel outputs.

4. The system of claim 1 , wherein each of the motion model kernels comprise a pixel mask that covers possible target movements in accordance with a respective motion model of each of the motion model kernels.

5. The system of claim 1 , wherein the processor is configured to fuse the plurality of kernel outputs via a max-minus-min fusion.

6. The system of claim 1 , wherein the fused estimate comprises estimated velocity for the pixel, wherein the estimated velocity is indicative of direction of movement of the target.

7. A method for tracking a target in images, the method comprising:

receiving, via an input module, a first change detection image corresponding to a first point in time and a second change detection image corresponding to a second point in time that is subsequent the first point in time, wherein the first change detection image and the second change detection image have been processed to remove static objects therefrom, wherein the first change detection image and the second change detection image have a plurality of pixels;

processing, via a processor, the first change detection image and the second change detection image, wherein processing the first change detection image and the second change detection image comprises generating a plurality of kernel outputs for a first pixel in the second change detection image via a plurality of different recursive motion model kernels, wherein the outputs are based upon the first pixel in the second change detection image and a second pixel in the first change detection image, wherein the plurality of different recursive motion model kernels are representative of a plurality of different motion models, and further wherein the outputs comprise estimated velocities and estimated energies for the first pixel; and

fusing, via the processor, the outputs to produce a fused estimate, wherein the fused estimate comprises a fused velocity estimate and a fused energy estimate that is based upon the estimated velocities and the estimated energies output by the plurality of different recursive motion model kernels, the fused estimate is representative of motion of the target between the first point in time and the second point in time.

8. The method of claim 7 , wherein fusing the outputs comprises determining an estimated velocity that is highest from amongst the estimated velocities, wherein the fused velocity estimate is the estimated velocity that is highest.

9. The method of claim 7 , wherein the processing of the received image in parallel via a plurality of different recursive motion model kernels comprises applying a pixel mask to the first change detection image and the second change detection image that covers possible movements of the target between the first point in time and the second point in time.

10. The method of claim 7 , wherein the fusing of the outputs to produce a fused output comprises fusing the outputs via one of a max-minus-min fusion or a velocity fusion.

11. The method of claim 7 , wherein the outputs comprise a directional energy vector for each recursive motion model kernel.

12. The method of claim 11 , wherein fusing the outputs to produce the fused estimate comprises computing an estimated directional energy vector, the estimated directional energy vector being a difference between an estimated energy in the outputs with a highest magnitude and an estimated energy in the outputs with a lowest magnitude.

13. A computer-readable device comprising instructions that, when executed by a processor, cause the processor to perform acts comprising:

receiving, via an input module, a first change detection image corresponding to a first point in time and a second change detection image corresponding to a second point in time that is subsequent the first point in time, wherein the first change detection image and the second change detection image have been processed to remove static objects therefrom, wherein the first change detection image and the second change detection image have a plurality of pixels;

processing, via a processor, the first change detection image and the second change detection image, wherein processing the first change detection image and the second change detection image comprises generating a plurality of kernel outputs for a first pixel in the second change detection image via a plurality of different recursive motion model kernels, wherein the outputs are based upon the first pixel in the second change detection image and a second pixel in the first change detection image, wherein the plurality of different recursive motion model kernels are representative of a plurality of different motion models, and further wherein the outputs comprise estimated velocities and estimated energies for the first pixel; and

fusing, via the processor, the outputs to produce a fused estimate, wherein the fused estimate comprises a fused velocity estimate and a fused energy estimate that is based upon the estimated velocities and the estimated energies output by the plurality of different recursive motion model kernels, the fused estimate is representative of motion of the target between the first point in time and the second point in time.

14. The computer-readable device of claim 13 , wherein fusing the outputs comprises determining an estimated velocity that is highest from amongst the estimated velocities, wherein the fused velocity estimate is the estimated velocity that is highest.

15. The computer-readable device of claim 13 , wherein the processing of the received image in parallel via a plurality of different recursive motion model kernels comprises applying a pixel mask to the first change detection image and the second change detection image that covers possible movements of the target between the first point in time and the second point in time.

16. The computer-readable device of claim 13 , wherein the fusing of the outputs to produce a fused output comprises fusing the outputs via one of a max-minus-min fusion or a velocity fusion.

17. The computer-readable device of claim 13 , wherein the outputs comprise a directional energy vector for each recursive motion model kernel.

18. The computer-readable device of claim 17 , wherein fusing the outputs to produce the fused estimate comprises computing an estimated directional energy vector, the estimated directional energy vector being a difference between an estimated energy in the outputs with a highest magnitude and an estimated energy in the outputs with a lowest magnitude.