Is it possible to track object in a 360 degree camera

Please provide complete information as applicable to your setup.

• Hardware Platform (Jetson )
• DeepStream Version 6.2
• Issue Type( questions,)
I have a 360 degree camera. If an object is moved around this camera in a circle, can DeepStream track the object and maintain its tracking ID?
Below is the view of the camera.

Yes, you can have a try with /opt/nvidia/deepstream/deepstream-6.3/sources/apps/sample_apps/deepstream-test2 with config_tracker_NvDCF_accuracy.yml.
But we can see your stitched video has mess when crossing camera. Maybe it is hard to maintain the same tracking ID.

Hi thank for your quick reply, I message you one video show when person cross the boundary of the image, the tracker could not keep the tracking ID. However, because this is a 360 degree camera(cylindrical view), so I would like keep the tracking ID even the person cross the image boundary. Or did I missing some config? I already use the tracker config config_tracker_NvDCF_accuracy.yml.

Can you have a try with DeepStream latest version 6.3?

I checked your video, the Person 0 and Person 2 have big difference. One is the front of the person, another is the back of the person. Is it possible to try to keep more similarity when cross the boundary?

The person in your video move out of the screen and enter the screen from the other side. So the IOU of the BBox is 0. The size of the BBox is changeing as the body has been partially covered sometimes in your video. So I tuned below parameter of nvtracker. It can keep the same ID when run deepstream-app with PeopleNet detector on DeepStream 6.3.

config_tracker_NvDeepSORT.yml-tuned (6.2 KB)

This configuration work!! Thank you very much.

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