Interesting project! You could run the neural networks directly on the pi. Expect 3 fps for object detection, try ncnn or deepdetect + ncnn for ease of use. Tha

Interesting project! You could run the neural networks directly on the pi. Expect 3 fps for object detection, try ncnn or deepdetect + ncnn for ease of use. That's without GPU. With a Jetson Nano expect up to 50 fps. You could detect more classes as well, as needed.