Overview
Neighborhood cats treating front-yard flower beds as personal litterboxes is a common frustration, and commercial repellent sprays are often entirely ineffective. To solve this without resorting to manual intervention, I engineered an automated, edge-AI deterrent system. By combining a low-light camera, a Raspberry Pi 5, and a custom-tuned YOLO computer vision model, this system actively monitors the yard, accurately identifies feline intruders, and triggers a physical lawn sprinkler via GPIO relays to humanely deter them.
Current Features
- Real-Time Feline Detection: Evaluates video feeds locally to identify cats while ignoring standard neighborhood traffic.
- Automated Data Pipeline: Includes a lightweight motion-detection script that continuously curates and formats custom training data.
- Advanced False-Positive Filtering:
- Utilizes class-specific confidence thresholds to avoid mistaking humans for cats.
- Implements bounding-box size limits to ignore static landscaping and bushes.
- Tracks spatial movement to ensure detections are actual moving targets.
- Hardware Actuation: Interfaces directly with physical relays and solenoid valves to trigger a garden hose sprinkler automatically.
Highlights & Outcomes
- Optimized Edge Inference: Achieved high-performance, local inference without cloud dependencies by leveraging the Raspberry Pi 26 TOPS AI HAT+.
- Robust Environmental Adaptation: Overcame the limitations of off-the-shelf COCO datasets by engineering a custom pipeline to fine-tune the model against specific, real-world yard conditions (including harsh shadows and infrared night vision).
- Weatherproof Deployment: Successfully designed and deployed a durable hardware enclosure with proper wiring protections to shield the electronics from moisture and the Albuquerque heat.
- Hardware/Software Integration: Bridged the gap between high-level Python machine learning scripts and low-level electronic engineering (actuating valves via GPIO pins).
Architecture
- Core Language: Python
- Vision Model: YOLO26m (Fine-tuned)
- Controller Hardware: Raspberry Pi 5 & 26 TOPS AI HAT+
- Optical Hardware: Arducam B0647 (STARVIS IMX290)
- Physical Actuation: 5V Relay, 12V Solenoid Valve, Standard Lawn Sprinkler