Dear NVidia-fellows,
I’d like to develop a prototype for overhead people detection and people tracking in malls. The functionality should basically be that it counts people reliably.
My hardware setup is an NVidia jetson nano.
I’ve taken a look at the pretrained models:
https://devblogs.nvidia.com/training-custom-pretrained-models-using-tlt/
Can you think of datasets available online that I can use for the training? Can you recommend a pretrained model to start with for overhead people detection?
The footage I have was recorded with a fisheye lens mounted almost directly above people at the entrace. So the people appear quite large despite the large field of view. Can I do something to help the models by e.g. placing the camera at a larger distance from the people?
Thank you!