Optimize CSRT tracker?

asked 2020-08-07 03:02:18 -0500

I'm new to OpenCV and was hoping that someone more experienced might be able to point me in the right direction. I'm doing motion tracking on black-and-white videos of insects in flight. Running CSRT does a decent job of tracking the insects much of the time, but it sometimes loses the target when the insect crosses a changing background at a distance. Since my target appearance is very specific, I was thinking that I might be able to train my own classifier and use it to improve the accuracy of the CSRT tracker. Is there a way to add a custom Haar cascade (or other classifier method) to the CSRT source code, or would it be standard practice to add a classifier on top of the CSRT (i.e. write a function to reacquire the target every n frames using the custom classifier), or is there some other way this is typically done? I appreciate any insight. Thanks!

edit retag flag offensive close merge delete


an image of the object you track will be useful

sturkmen gravatar imagesturkmen ( 2020-08-07 04:55:52 -0500 )edit

also, can you try to highlight the weakness of the CSRT tracker in your case ?

berak gravatar imageberak ( 2020-08-07 08:08:12 -0500 )edit

Another option is tracking by detection, where you run your insect detector every X frames and associate new detections to established tracks or initialize new tracks. This is a more robust approach than giving a tracker a single detection at frame 0 and never running the detector again. How are you detecting the insects?

Der Luftmensch gravatar imageDer Luftmensch ( 2020-08-07 09:57:47 -0500 )edit

From now on I will downvote and then delete abandoned questions like this

sturkmen gravatar imagesturkmen ( 2020-08-17 08:42:53 -0500 )edit