Foreground motion detection in compressed video data

1. A system to distinguish foreground motion from background motion in video data, the system comprising:

a memory configured to store compressed video data; and

a processor block configured to:

select, from the compressed video data, a block coding unit (BCU) that potentially includes the foreground motion;

determine, without decompression of the BCU, a statistic based on one or more of:

a transform-coefficient-magnitude sum of the BCU,

a motion-vector-magnitude sum of the BCU,

a background mean associated with the compressed video data, and

a background co-variance associated with the compressed video data; and

determine, based on a comparison of the statistic to a threshold, whether the BCU includes the foreground motion.

2. The system of claim 1 , wherein the processor block is further configured to determine the transform-coefficient-magnitude sum based on: at least one quantized transform coefficient of the BCU; and at least one corresponding quantization step size.

3. The system of claim 2 , wherein the processor block is further configured to determine the transform-coefficient-magnitude sum based on a number of bits consumed in coding the at least one quantized transform coefficient of the BCU.

4. The system of claim 1 , wherein the processor block is further configured to determine the motion-vector-magnitude sum based on:

at least one BCU partition/prediction unit associated with a motion vector of the BCU; and

a reference frame index associated with the motion vector of the BCU.

5. The system of claim 1 , wherein the processor block is further configured to compute a Mahalanobis distance between the statistic and the background mean to determine whether the BCU includes the foreground motion.

6. The system of claim 1 , wherein the processor block is further configured to:

determine whether a BCU pixel is similar to a co-located pixel in a background frame; and

in response to the determination that the BCU pixel is similar to the co-located pixel, determine that the BCU pixel is a background pixel.

7. The system of claim 1 , wherein the processor block is further configured to update the threshold based on at least one of a confidence interval and a number of foreground pixels.

8. The system of claim 1 , wherein the processor block is further configured to:

generate a temporal median based on a subset of the compressed video data; and

generate the background mean and the background co-variance based on the temporal median.

9. A method to identify foreground motion in compressed video data, the method comprising:

selecting, from the compressed video data, at least one block coding unit (BCU);

determining, without decompressing the at least one BCU: a transform-coefficient-magnitude sum of the at least one BCU, and a motion-vector-magnitude sum of the at least one BCU;

generating a background mean and a background co-variance based on a sliding temporal window with varying weights;

determining a statistic based on the transform-coefficient-magnitude sum, the motion-vector-magnitude sum, the background mean, and the background co-variance;

comparing the statistic to a threshold;

in response to a determination that the statistic is above the threshold, determining that the at least one BCU includes the foreground motion; and

in response to a determination that the statistic is below the threshold, determining that the at least one BCU does not include the foreground motion.

10. The method of claim 9 , wherein determining the transform-coefficient-magnitude sum is based on at least one of:

a quantized transform coefficient of the at least one BCU;

a quantization step size corresponding to the quantized transform coefficient; and

a number of bits consumed in coding the quantized transform coefficient.

11. The method of claim 9 , wherein determining the motion-vector-magnitude sum is based on:

at least one BCU partition/prediction unit associated with a motion vector of the at least one BCU; and

a reference frame index associated with the motion vector.

12. The method of claim 9 , further comprising:

in response to the determination that the at least one BCU includes the foreground motion, determining whether a first pixel in the at least one BCU is similar to a co-located second pixel in a background frame; and

in response to the determination that the first pixel is similar to the co-located second pixel, determining that the first pixel is a background pixel.

13. The method of claim 9 , further comprising:

generating a temporal median based on a subset of the compressed video data; and

generating the background mean and the background co-variance based on the temporal median.

14. The method of claim 9 , wherein generating the background mean and the background co-variance further comprises generating the background mean and the background co-variance using a recursive process.

15. A system to identify foreground motion in video data, the system comprising:

a memory configured to store compressed video data; and

a processor block configured to:

select, from the compressed video data, a block coding unit (BCU) that potentially includes the foreground motion;

determine, without decompression of the BCU: a transform-coefficient-magnitude sum of the BCU, and a motion-vector-magnitude sum of the BCU;

generate a background mean and a background co-variance based on a sliding temporal window with varying weights;

compare the background mean and the background co-variance to the transform-coefficient-magnitude sum of the BCU and the motion-vector-magnitude sum of the BCU;

determine, based on the background mean, the background co-variance, the transform-coefficient-magnitude sum, and the motion-vector-magnitude sum, that the BCU includes the foreground motion.

16. The system of claim 15 , wherein the processor block is further configured to determine the motion-vector-magnitude sum of the BCU based on:

at least one BCU partition/prediction unit associated with a motion vector of the BCU; and

a reference frame index associated with the motion vector.

17. The system of claim 15 , wherein the processor block is further configured to:

compute a Mahalanobis distance between the background mean and a combination of the transform-coefficient-magnitude sum of the BCU and the motion-vector-magnitude sum of the BCU; and

determine that the BCU includes the foreground motion based on the Mahalanobis distance.

18. The system of claim 15 , wherein the processor block is further configured to:

determine whether a first pixel in the BCU is similar to a co-located second pixel in a background frame; and

in response to the determination that the first pixel is similar to the co-located second pixel, determine that the first pixel is a background pixel.