Volumetric sampling with correlative characterization for dense estimation

1. An apparatus for optical flow estimation, comprising:

a memory configured to store data corresponding to one or more frames; and

a processor coupled to the memory and configured to:

determine an optical flow prediction associated with a plurality of frames;

determine a position of a feature associated with a first frame;

determine, based on the position of the feature in the first frame and the optical flow prediction, a position estimate of a search area for searching for the feature in a second frame;

generate a correlation volume comprising the search area; and

determine, from within the correlation volume, a position of the feature in the second frame.

2. The apparatus of claim 1 , wherein determining the position estimate of the search area comprises offsetting the position of the feature associated with the first frame by an optical flow prediction value of the optical flow prediction associated with the feature.

3. The apparatus of claim 2 , wherein the optical flow prediction value associated with the feature is determined based at least in part on the position of a pixel associated with the feature in the first frame.

4. The apparatus of claim 1 , the processor configured to:

perform object detection on the feature to determine an object type associated with the feature; and

determine the optical flow prediction based on the object type associated with the feature.

5. The apparatus of claim 1 , wherein a plurality of features associated with the first frame comprises respective features corresponding to pixels of the first frame.

6. The apparatus of claim 1 , wherein the feature associated with the first frame comprises a pixel location of at least a portion of an object in the first frame.

7. The apparatus of claim 6 , wherein determining the optical flow prediction comprises obtaining an optical flow vector corresponding the pixel location of at least the portion of the object.

8. The apparatus of claim 1 , wherein the search area includes at least an estimated pixel position of the feature in the second frame.

9. The apparatus of claim 8 , wherein the estimated pixel position is centered within the search area.

10. The apparatus of claim 8 , wherein the estimated pixel position is off-center with respect to the search area.

11. The apparatus of claim 1 , wherein the search area comprises a rectangle having a height and a width, wherein the height and width are different.

12. The apparatus of claim 1 , wherein the first frame precedes the second frame in a frame sequence.

13. The apparatus of claim 1 , wherein the processor is configured to:

for each feature of a plurality of features in the first frame, determine a corresponding estimated position of the plurality of features in the second frame.

14. The apparatus of claim 1 , wherein the optical flow prediction comprises a flow target map.

15. The apparatus of claim 1 , wherein the optical flow prediction corresponding to the feature is based on at least a third frame, wherein the third frame precedes the first frame and the second frame in a frame sequence.

16. A method of optical flow estimation, comprising:

determining an optical flow prediction associated with a plurality of frames;

determining a position of a feature associated with a first frame;

determining, based on the position of the feature in the first frame and the optical flow prediction, a position estimate of a search area for searching for the feature in a second frame;

generating a correlation volume comprising the search area; and

determining, from within the correlation volume, a position of the feature in the second frame.

17. The method of claim 16 , wherein determining the position estimate of the search area comprises offsetting the position of the feature associated with the first frame by an optical flow prediction value of the optical flow prediction associated with the feature.

18. The method of claim 17 , wherein the optical flow prediction value associated with the feature is determined based at least in part on the position of a pixel associated with the feature in the first frame.

19. The method of claim 16 , further comprising:

performing object detection on the feature to determine an object type associated with the feature; and

determining the optical flow prediction based on the object type associated with the feature.

20. The method of claim 16 , wherein a plurality of features associated with the first frame comprises respective features corresponding to pixels of the first frame.

21. The method of claim 16 , wherein the feature associated with the first frame comprises a pixel location of at least a portion of an object in the first frame.

22. The method of claim 21 , wherein determining the optical flow prediction comprises obtaining an optical flow vector corresponding the pixel location of at least the portion of the object.

23. The method of claim 16 , wherein the search area includes at least an estimated pixel position of the feature in the second frame.

24. The method of claim 23 , wherein the estimated pixel position is centered within the search area.

25. The method of claim 23 , wherein the estimated pixel position is off-center with respect to the search area.

26. The method of claim 16 , wherein the search area comprises a rectangle having a height and a width, wherein the height and width are different.

27. The method of claim 16 , wherein the first frame precedes the second frame in a frame sequence.

28. The method of claim 16 , further comprising:

for each feature of a plurality of features in the first frame, determine a corresponding estimated position of the plurality of features in the second frame.

29. The method of claim 16 , wherein the optical flow prediction comprises a flow target map.

30. The method of claim 16 , wherein the optical flow prediction corresponding to the feature is based on at least a third frame, wherein the third frame precedes the first frame and the second frame in a frame sequence.