I need some help on solving this issue. Any tips are appreciated.

The goal of the project is to try to track a Bounding Box object, chosen arbitrarily in an image, throughout many frames of a recorded video.

I extract from the chosen bounding box cv2.goodFeaturesToTrack and use a sparse optical tracker cv2.OpticalFlowPyrLK in OpenCV to track these features. However I need to also update the bounding box corners for each frame.

Since the bounding box corners are initially chosen to be random and do not necessarily belong with the features in the same rigid object , I am looking for using the information on the movement of these special features to infer the new bounding box coordinates.

I tried to add the bounding box corners to the tracker but it is not robust and loses it in a few frames.

I am using OpenCV in python.

Thank you.

  • $\begingroup$ My perception is that the question would benefit from a bit more focus. The updating of the bounding box once the area is identified is an extremely simple step. Are there other conditions during the identification of the region of interest that are challenging? (You might also be interested in normalised cross correlation ) $\endgroup$
    – A_A
    Feb 20, 2020 at 10:47
  • $\begingroup$ Thank you @A_A. Only the special features in the bounding box are tracked throughout the frames and not the box itself. My question focuses on using the update on the sparse special features to also update the bounding box corners. So I choose a bounding box, I pick good features in it and track these features only and right now cannot find a way to update the bounding box coordinates. $\endgroup$ Feb 27, 2020 at 12:18


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge that you have read and understand our privacy policy and code of conduct.