Catcam Mouse Detector Computer Vision Project

We have a cat, called Nala, who loves us so much that she continuously brings us presents. We have had any number of mice - of all varieties and sizes you can imagine. We have had rats. We have had even a Sand Marten. She will chase and catch anything. Some of these she eats, because clearly there is nothing tastier than a nice fresh rodent, but many she brings home - either in bits or fully alive, or somewhere inbetween. Clearly something has to be done.

As an emergency measure we simply close the catflap at night but we live at 500m altitude and it gets fairly cold here in winter - this calls for technology. I have wrapped the catflap in a wooden tunnel that guides the cats past a Raspberry Pi 4 with a camera on it and from this tunnel I get a reasonably exposed set of videos mostly in IR illumination, not too blurry if the cats move quickly, with a decent image of what the cat has in its mouth. Getting a decent image quality was quite tricky and time consuming, required several experiments.

Behind this is a simple piece of Python that just streams the video on my LAN so I can develop on my desktop. A client on the stream records videos every time motion is detected in a window on the frame. Currently the Pi just runs the streaming.

The code is on GitHub, currently private - when I take out some of the hacks I will make it public but it is based on the standard Pi Streaming implementation and the OpenCV standard motion detection algorithm.

This project has the ML part of the work - I use VLC to make still image snapshots from the video, this is comfortable. The rest churns through Roboflow using the standard features.