Cowcatcher AI
Open source project
Combining the Cowcatcher vision model with the AI-detector creates an automated alert system that monitors cows to detect "mounting" behavior.
The Basics
- 24/7 Monitoring: The program monitors a live camera feed pointed at the cows around the clock.
- Artificial Intelligence: It uses AI to automatically recognize when one cow mounts another, built on Ultralytics Yolo.
- Instant Alerts: Immediately sends photo and video notifications to your phone via Telegram or Home Assistant.
- Non-Intrusive: Your cameras remain functional; we simply use an additional stream.
- Local Processing: Everything runs locally on your own computer in the stable or office.
The "Smart" Part
Learned Behavior: This program has been trained to recognize heat behavior using thousands of images.
Confidence Score: The system assigns a reliability score to every detection.
How it works in practice
- Continuous Monitoring: The camera films the cows all day.
- Smart Analysis: The program analyzes activity every few seconds.
- Detection: High certainty detections trigger image and video capture.
- The Perfect Shot: It collects several seconds of footage for the clearest view.
- Sending Notifications: You receive a "Heat Detected" alert with visual proof.
Important Considerations – You Stay in Control
A Tool, Not a Replacement: AI can trigger false alarms; you make the final decision.
Animal Identification: You must identify the specific cow from the photo yourself.
Physical Checks Remain Vital: Always check for mucus and physical signs when in doubt.
An Extra Pair of Eyes: A digital assistant for night-time or when you are away.
Practical Workflow: Receive Notification →
View Photo → Identify Animal → Check Physically → Take Decision
Why This System is Valuable
- No need to watch the cows constantly.
- Don't miss heat cycles at night.
- Immediate alerts allow for fast action.
- Identify the exact onset of heat for precision insemination.
Data & Privacy
- Local Hosting: Runs on your own computer without cloud connections.
- Requirements: Computer must stay on and connected to the internet for alerts.
- Platforms: Uses Telegram or Home Assistant for local setups.
- Storage: Images are stored locally; only the best are sent to your phone.
- Continuous Improvement: Option to further train the program yourself.
What You Need
- A Computer: Ideally with an NVIDIA GPU (GTX 1000-series or newer).
- IP Camera: Most RTSP-capable cameras work (e.g., Reolink Trackmix).
- POE Switch: To power cameras via a single cable.
- LAN Cable: Approximately €1 per meter.
- Internet Connection.
What We Provide
- The Trained AI Model: The neural network 'brain'.
- Privacy: Contains no images from other farms.
- Open Source Software: A reusable third-party AI detector.



