Tracking face and body

[sevencong]

Can we tracking detected face or/and body movement in multiple sequence frames in-order to GROUP detected faces. Then we do not need to match them later if we found in database.

example:
frame 1: [face group id1, face group id 2], -> bad quality, do .match() and not found in database
frame 2: [face group id1, face group id 2] -> bad quality, do .match() and not found in database
frame 3: [face group id1-nameX, face group id 2-nameY] -> GOOD quality, do .match() and found in database
frame 4: [face group id1-nameX, face group id 2-nameY] -> not need to .match() again
frame 5: [face group id1-nameX, face group id 2-nameY] -> not need to .match() again

[vladmandic]

I'll think about it a bit more, but not sure how would that be any faster than using match explicitly as library would have to do the same thing anyhow.

[sevencong]

at some situations, .match() return low similarity on same face (different angle...), if we have them grouped (tracking), we have more information on faces matching.

[vladmandic]

question is how to group them as detection order depends on size and location within frame and that can change in video input.

[sevencong]

I found python code for tracking https://github.com/abewley/sort

[vladmandic]

I'll take a look