How I can use the jetson nano for dangerous people around school

Hi, everybody,
I’m doing a project. In this project, I want to classify the people around the building as dangerous or not. I want to use a Jetson nano and a raspberry camera. How can I follow this path? Which model should I use for high fps and consistent results?

There are a lot of options to detect person, e.g. with OpenCV: Real-time Human Detection with OpenCV

I wonder what your definition of „dangerous people“ is?

for example if a person is waiting around the building for a long time or has a gun

For the former you will need a detector that can re-identify persons/faces over a time period, the latter might be covered by Yolo or SSD-Mobilenet.
A good starting point might be one of the model zoos:
Open Model Zoo Demos - OpenVINO™ Toolkit / GitHub - openvinotoolkit/open_model_zoo: Pre-trained Deep Learning models and demos (high quality and extremely fast)