parameters of Kalman filter

[abewley]

parameters of Kalman filter

[abewley]

The parameters were selected from a few samples on a subset of the MOT 2015 training dataset.
The problem you are experiencing is due to the initial detection not knowing the velocity of the object so it will lag if the object is moving at the point of the first detection. You can try increasing the uncertainty (line 90) or work out a way to measure velocity (i.e. optical flow) with each detection.

[landiaokafeiyan]

Thanks Abewley !
Actually I found the lag mainly because the RGB image is late then the depth image. but the parameters also contribute to the results.
Is that means we need to compute the velocity for every objects?because they may have different velocities(for example pedestrian and car), that will need heavy computational.
for that case what we can do to set the parameters?

[abewley]

Not sure where you are getting the "depth image" from but it sounds like you are doing more than 2D image tracking. Yes, estimating the velocity for each object would require additional components not supplied in this repo.

[landiaokafeiyan]

Yeap ,I am doing 3D image tracking. Many thanks for your reply!