Hi everyone,

I am working on a project where I am using an OAK-D pro camera. I am using the camera to de object detection and localization using YOLOv8n. After this I am taking the location with respect to the camera, and publishing this to a ROS node in order to later annotate a map being created by the host computer.

However, I ran into an issue. When we go over a place we have already been, we are publishing the same object multiple times to our ROS node. I want to be able to recognize an object I have already seen so I don't publish its location more than one time.

I tried using the built in ObjectTracker node. But it is only able to track objects when in frame, once out of frame it loses them and gives them a new instance ID when they reappear on screen. What solutions are there for my problem that I can implement. Are these solutions capable of running locally on the OAK-D device? Or do they have to be implemented on the host computer?

    isagmz
    At the moment, you will have to do this on host side. The implementation of ObjectTracker currently only works until the object is deleted when lost.
    You would need to implement some kind of reindentification system based on detected objects to do that.

    Thanks,
    Jaka