Hello everyone. Here is the deal:
I have a physical device that can track objects according to their location on the screen. the location is given by YOLO detection.
As far a I know, YOLO is not dedicated for tracking. That’s why it uses external trackers such as BoT-Sort and bytetrack.
If I’m giving my physical device the location of the tracked object based on the center of the bounding box it is very noisy because the size of the bounding box constantly changes.
Is it possible to get the location of the object directly from the tracker?
If not, I will appreciate it if you suggest me some methods to reduce the noise (Other than moving average )
Why not moving average (assume you mean unweighted moving average)? Have you tried moving average smoothing? Which ones have you tried and of those, why didn’t they work?
See the tracking source code if you’d like to extract something directly, but my guess is that’s more complex than what you truly need.
From what I see in the tracker source code, the tracker throws me back to the bounding box so it doesn’t help. However, I did not try moving average filter yet so it might work. Tnx!
Usually tracking is made smooth by applying kalman filter. But that’s already integrated into the tracker. The other type of filter that can help is the 1€ filter.
)