I'm using YoloSpatialDetectionNetwork which forwards detections into ObjectTracker. I'm working on people counting project where camera is located above the entrance on the ceiling and people are going in and out.
I'm experiencing surprisingly often the situation, that one person leaves the area directing up on the frame and different person enters almost at the same time from up directing down and tracker thinks, it is the same object and keeps tracking different person under the same ID. I'm using now "ZERO_TERM_COLOR_HISTOGRAM" tracker type.
Can you advice some approach I can avoid this "false reidentifiactions"?
I have in mind workaround: Count EVERY crossing as a new person entering or leaving even if it will be single ID. But there can be situation, where the same person really turns and goes back where she came from. So It can lead again to inaccurate counting.