Introduction
Main server detects eating cat on received images and persists events to disk.
API call is used to analyze time periods and remove false positives.
Limitations
Would not work at night time by default - requires infrared camera.
If you could place camera at right spot, then you could reduce false detections to zero by computing distance between cat and bowls, but in our camera setup it was not possible.
Project installation
- Create Python 3.10 environment using conda /
- Install dependencies with "pip install -r requirements.txt"
Server
Run main application with "uvicorn main:app --host 0.0.0.0 --port 8082".
Configured to use YOLO v5 large model by default.
Client application
Run "python camera-app.py 8082".
Requires DepthAI camera, but could be easily adapted to use any web camera that allows capturing of RGB image 1 time per second.
Helper application
Run "python show-detections.py " to analyze given image you show detected bounding boxes, classes and confidence levels.