catcam

RTSP camera + YOLO cat detection. When a cat is in a configured zone, it can snapshot, record a clip, and play a deterrent on a Sonos speaker.

Setup

Python 3.11+ recommended.

python3 -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt

Copy the example config and edit it:

cp config.example.yaml config.yaml

config.yaml is gitignored. Set at least:

• rtsp_url — your camera stream
• model.path — ONNX weights (see below)
• audio.sonos_ip — if you want the deterrent

Model

The scripts load an ONNX file. Export one from a YOLO checkpoint:

yolo export model=yolo11n.pt format=onnx imgsz=640 opset=12

Point model.path at the resulting .onnx file.

The example config uses stock COCO class ids (person 0, cat 15, dog 16). If you train a custom model, update classes and dataset.coco_to_dataset to match.

Camera

Put the full RTSP URL in rtsp_url. Prefer the camera's main stream. To check the feed without running detection:

python test_stream.py

That writes frame.jpg if it can read frames.

Run

python catcam.py

Open http://<this-machine>:8080 for the live view. Draw zone polygons there, then paste the coordinates into config.yaml under zones: and restart. With empty zones, the whole frame is active.

Optional: put deterrent.wav next to the scripts (or change audio.file) so Sonos can play it.