c++ - How can we use both Hamming distance and distance between coordinates to match features? -

as known, tracking objects in opencv can use:

• featuredetector find features
• descriptormatcher match similarity between features of desired object , features of current frame in video
• and use findhomography find new position of object

for matching features descriptormatcher uses hamming distance (value of difference between 2 sequences of same size, not distance between coordinates).

i.e. find similar object in current frame, not nearest previous position (if know it).

how can use match both hamming distance , distance between coordinates, example, given weight of both, not hamming distance?

it solve following problems:

• if start track object position (x,y) on previous frame, , current frame contains 2 similar objects, find similar, not nearest. due inertia coordinates changes slower similarity (a sharp change in light or rotation of object). , must find similar object nearest coordinates.

• thus find features, not similar, give accurate homography, because exclude features, which, although similar, far away in coordinates , belong other objects.

what need like:

1. compute matches usual.
2. dmatch has queryidx , trainidx indices. can use these retrieve corresponding keypoints. compute euclidean distance between them, , update value distance if dmatch kind of weighting function.
3. sort matches distance (since distance has changed).

now matches vector sorted according both hamming distance between descriptors , euclidean distance between keypoints.