I am using Opencv (Python) default trackers in an object detection and tracking task. As far as I know, we need to initialize some bounding boxes in order to make our trackers work so a detection algorithm will handle this task for us. The problem is if one (or even more) object was not detected in the initializing phase, the trackers will have to deal with it. The simplest solution is to perform the detection algorithm several times through frames to make sure that all objects are included, which is probably very computationally expensive. I wonder if there exists an effective way to deal with this problem?
1 Answers
Most people do it manually on the first frame using the cv2.selectROI() method, which returns a bounding box. This is described in some detail here:
https://www.learnopencv.com/multitracker-multiple-object-tracking-using-opencv-c-python/
For cases where that is not practical (e.g., thousands of objects), you can automate the object detection part, and just make sure you find a good frame to start with that gets things right.
Related