An automatic front doorbell camera with an alert when someone or something is at the door can be a helpful tool. A low-tech option with a Raspberry camera and low-footprint desktop software running in the background can be an ideal solution.
The task was to provide an automatic detector of humans on low-quality video feeds that can run with low computational power.
We provided an optimized algorithm that can run on a single-board computer taking advantage of the Yolov5n6 detector and OpenCV for decoding the RTSP video feed. The detected object is then saved, and we use Plyer to display notifications on the user’s desktop.