Struggling to track high speed objects with DeepStream


I have been trying to get the tracker to work well on small objects at a very high speed where the object is on screen for around 12 frames, and have been facing problems of:

  • Bounding box drifting - Initially the bbox would track the object well but over the next frames, the bbox would lag behind and only cover half the object

  • Re-ID issue - Whereby the same tracker ID would be given to a different object of the same class

Looking at the parameters for the tracker, it seems like turning off the state estimator (set to 0) seems to stop the bbox drifting issue, but would it affect the overall performance?

Also looking at the description of the state estimator, if it was set to 1 (Simple KF) with 6 states, is there any parameters whereby I can set the dy value? As I know which direction the object would be travelling in. To prevent any tracker IDs from ever going in the wrong direction

• Hardware Platform (Jetson / GPU) Jetson Orin AGX
• DeepStream Version 6.2
• JetPack Version (valid for Jetson only) 5.1-b147
• TensorRT Version
• Issue Type( questions, new requirements, bugs) Question
• How to reproduce the issue ? (This is for bugs. Including which sample app is using, the configuration files content, the command line used and other details for reproducing)

  • Tracker: nvdcf
  • Model: yolov4, cspdarknet 19
  • Video speed: 30fps
  • Video resolution: 1920 x 1080
  • Size of objects: 67 x 67

There is no update from you for a period, assuming this is not an issue anymore. Hence we are closing this topic. If need further support, please open a new one. Thanks

Can you video stream to reproduce the issue? So we can check and give advice.

Do you see the issue when using NvDeepSORT tracker?

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