Approximate cross-check for real-time feature matching

1. A method of feature matching comprising the steps of:

determining a set of correspondences between a set of reference feature points for a first image and a set of target feature points for a second image using a processor to compute a distance function for each pair of reference and target feature points; and

reducing the set of correspondences while the set of correspondences is being determined using an approximate cross-check to obtain a set of valid correspondences, wherein said processor determines whether a valid correspondence exists for a respective reference feature point after said distance function has been computed for all pairs of feature points containing the respective reference feature point and continuously updates the set of valid correspondences found after the approximate cross-check.

2. The method according to claim 1 , wherein the determining and reducing steps are performed concurrently.

3. The method according to claim 1 , wherein the determining and reducing steps are configured to generate a set of arrays identifying the valid correspondences during a single pass through a set of possible pairs of said reference feature points and said target feature points.

4. The method according to claim 1 , wherein said feature matching with approximate cross-check is performed in real time.

5. The method according to claim 1 , wherein:

the set of valid correspondences found after the approximate cross-check are identified using a pair of arrays and the processor continuously updates the pair of arrays identifying the valid correspondences.

6. The method according to claim 1 , wherein said distance function compares two corresponding descriptors of each pair of reference and target feature points.

7. The method according to claim 1 , wherein:

for binary descriptors said distance function computes a hamming distance; and

for real descriptors said distance function computes at least one of an L1 norm, an L2 norm, or a combination thereof.

8. The method according to claim 1 , wherein said first image and said second image are frames in a video sequence.

9. The method according to claim 1 , wherein said first image and said second image are pictures of the same scene taken at different points in time.

10. An apparatus comprising:

a memory interface configured to perform read and write operations on a memory; and

a processor configured to (i) determine a set of correspondences between a set of reference feature points for a first image and a set of target feature points for a second image by computing a distance function for each pair of reference and target feature points, and (ii) reduce the set of correspondences while the set of correspondences is being determined using an approximated cross-check scheme to obtain a set of valid correspondences by determining whether a valid correspondence exists for a respective reference feature point after said distance function has been computed for all pairs of feature points containing the respective reference feature point and continuously updating the set of valid correspondences found after the approximate cross-check.

11. The apparatus according to claim 10 , wherein said processor determines and reduces the set of correspondences concurrently.

12. The apparatus according to claim 10 , wherein said processor is further configured to generate a set of arrays identifying the valid correspondences using a single pass through a set of possible pairs of said reference feature points and said target feature points.

13. The apparatus according to claim 10 , wherein said processor is further configured to:

continuously update a pair of arrays identifying the valid correspondences found after the approximate cross-check.

14. The apparatus according to claim 10 , wherein said distance function compares two corresponding descriptors of each pair of reference and target feature points.

15. The apparatus according to claim 10 , wherein:

for binary descriptors said distance function computes a hamming distance; and

for real descriptors said distance function computes at least one of an L1 norm, an L2 norm, or a combination thereof.

16. The apparatus according to claim 15 , wherein said distance function for binary descriptors is computed using a circuit configured to generate a vector through a bitwise exclusive-OR of two corresponding descriptors of each pair of reference and target feature points and sum across the vector.

17. The apparatus according to claim 10 , wherein said first image and said second image are at least one of frames in a video sequence or pictures of the same scene taken at different points in time.

18. The apparatus according to claim 10 , wherein said memory interface and said processor are implemented on a single integrated circuit.

19. The apparatus according to claim 10 , wherein said memory interface and said processor are part of a system on chip.

20. The apparatus according to claim 10 , wherein said memory interface and said processor are part of a processor chip of digital camera.