People detection in top down view

I would like to set up a pipeline to track people (more specifically their head) in a room and consider using a ZED camera to achieve that. Since I don’t have access to a ZED camera to test this, I would be happy to learn about experiences related to that beforehand.

Does anyone have experience with this? Does the object detector detect people reliably when the ZED camera is mounted nadir (on the ceiling looking down)? To not just track the center of the body, but the head, I would also apply the body tracking. Will this work when using a top down view?

Hi @thomasm

We have a dedicated model trained for head detection, so I don’t advise going for skeleton tracking from top down, the results will be not good at best.
That being said, I’m sorry to say we’ve not tested this workflow much internally. I’ll try to run some quick tests and I’ll get back to you.

Hi again @thomasm, sorry for the delay.

After some tests, it’s actually the opposite of what I said, sorry. The skeleton tracking is not the best, but a lot better than the head tracking in a top-down setup.
When the subject is just below the camera, the tracking is pretty much always lost, though.

Hi @JPlou,

thank you so much for looking into this and getting back to me so swiftly. Since for my application I expect to have people directly underneath the camera, I will have to look into other means to achieve my goal.