Revision history [back]

Hello,

I have some questions to ask regarding multiple camera tracking.

In an ideal scenario, if I have 2 cameras tracking multiple people and I project the coordinates of their feet onto a "top view" of the scene via the homography matrix, their coordinates should now be similar. But in practice, due to calibration errors, their coordinates will not be similar.

How then should I give a consistent label to all pairs of the "projected people" ? Do I just compute the euclidean distance from each point to all other points in the top view and assign the pair with the smallest distance the same ID if it is below a threshold ? Or do I use the epipolar constraint ?

Thanks.

Hello,

I have some questions to ask regarding multiple camera tracking.

In an ideal scenario, if I have 2 cameras tracking multiple people and I project the coordinates of their feet onto a "top view" of the scene via the homography matrix, their coordinates every pair of object corresponding to the same person should now be similar. have a similar coordinate. But in practice, due to calibration errors, their coordinates will not be similar.

How then should I give a consistent label to all pairs of the "projected people" ? Do I just compute the euclidean distance from each point to all other points in the top view and assign the pair with the smallest distance the same ID if it is below a threshold ? Or do I use the epipolar constraint ?

Thanks.

Hello,

I have some questions to ask regarding multiple camera tracking.

In an ideal scenario, if I have 2 cameras tracking multiple people and I project the coordinates of their feet onto a "top view" of the scene via the homography matrix, every pair of object corresponding to the same person should now have a similar coordinate. But in practice, due to calibration errors, their coordinates will not be similar.

How then should I give a consistent label to all pairs of the "projected people" ? Do I just compute the euclidean distance from each point to all other points in the top view and assign the pair with the smallest distance the same ID if it is below a threshold ? Or do I use the epipolar constraint is there a more efficient method ?

Thanks.