DetectNet is working but what about tracking an object that has been detected?


I have DetectNet working and detecting my objects ok as per dusty_nv demo. I only need to detect one type of object but multiple instances of that object. DetectNet works well for this.

What I need is to be able to track those objects and determine if they are getting closer to the camera and ideally approximate a distance. It is important for me to identify if the object is moving closer or further away and to differentiate between a newly detected object in the scene and one that is already in the scene, the actual tracking of the detected object.

Is DetectNet really what I want? Or is there a hybrid approach that I need, say with openCV tracking the detected object from DetectNet?

Can anyone shed some light on the right way to approach this?


DetectNet doesn’t detect distance, but openCV has support for distance estimation, you’ll need stereo cameras though.
Another option is getting an IR sensor to estimate distance.


You can find our Deepstream SDK for the object tracker.


Thank you for your help.