DeepSort not tracking when Detector lost an object

[OleksiiYeromenko]

I've compared results on the same video with pure Yolov5 and combined Yolov5 + DeepSort - results are the same. Whenever Yolov5 lost an object behind an obstacle, combined model also does not output object bounding boxes. Maybe I'm doing something wrong? Why DeepSort (Kalman filter) does not predict the next position of an object? How can I use DeepSort to improve detector results of a single object tracking?

[xuaner11111]

hello, do you solve this problem, i have the same problem with you

[dylantzx]

Hello @OleksiiYeromenko @xuaner11111, I am not an expert at this either but I think that to predict the next position of the object, you would have to call tracker.predict() only without tracker.update(detections) since there are no detections.

You can try tuning the std_weight_position and _std_weight_velocity in _kalman_filter.py to improve the kalman filter prediction results, or play around with the max_age and n_init in tracker.py too