xivo
xivo copied to clipboard
Improve feature tracking stability
Basic pyramid KLT is used for feature tracking. Due to challenging illumination change, the current feature tracker is only able to track features for a short period of time. Need to improve the stability of the feature tracker and make the life time of each feature longer.
I implemented the Farneback Optical Flow algorithm in the more_trackers
branch yesterday. My comments are:
- It's really slow compared to the basic pyramid KLT.
- It's very inaccurate for wide FOV cameras with lots of distortion.
For the sake of speed, the tracker really should be sparse, not dense. The only sparse tracker implemented in OpenCV is the basic pyramid KLT. The only other option I can think of at the moment is to just do feature matching instead of optical flow.