Missed object (False negative) tracking

Does “nvtracker” return Missed object bbox? I am using Python binding. If yes, how to enable it? What I seen in default config, it is just doing association and assigning ID to tracked object. I also read in doc (Gst-nvtracker — DeepStream 6.2 Release documentation) that it is able to assign ID even detector miss to detects some object in few in-between frames . but my question is whether it return that in-between bbox calculated by tracker.

Please refer this chapter which set detector interval to 2 with NvDCF tracker: Gst-nvtracker — DeepStream 6.2 Release documentation

1 Like

I have gone through the documentation that you suggested. I would like to clarify my previous statement. My solution does not require setting an interval, etc. Let me rephrase my previous message. I am interested in performing model inference on each frame, while also allowing the tracker to add bounding boxes that were not detected by the model inference. For example, suppose an object such as a car is initially detected by the model, but is then missed for a brief period of time before being detected again. In such cases, the tracker can still function effectively and assign the same ID to the initial and later bounding boxes (referred to as Shadow Tracking in the documentation). My objective is to obtain new bounding boxes from the tracker in situations where the model inference fails to detect them.

If the PGIE failed to detect the object, nvtracker will track it in Shadow Tracking mode. nvtracker don’t report the object during shadow tracking. Please refer: NvDCF Tracking between intervals - #10 by pshin

1 Like

This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.