People counting with computer vision on a Jetson Nano

Cameras watch a work area. The system counts people as they enter and leave, and checks whether each workstation is occupied, in real time.

I fine-tuned a pre-trained detection model with TensorFlow on a dataset labeled for this site. To make labeling fast, I built a small PyQt5 tool around LabelImg. The model was then converted and optimized to run on a Jetson Nano next to the cameras, with OpenCV handling the video streams.