Summary
The system employs twelve high‑definition cameras, including infrared sensors, to capture video day and night in a farm setting.
Using a computer vision model and a tracking algorithm, the system identifies four key behaviours—feeding, standing, lying, and mounting—and assigns each animal a unique ID.
Performance tests showed a mean average precision of 99.5% and a recall of 99.6%, processing about 31 frames per second on an NVIDIA RTX 3060 graphics card.
Continuous behavioural monitoring can alert farmers to health, comfort, or reproductive changes, reducing the need for constant human observation.
The technology is still experimental, tested on a single farm, and aims to expand to more behaviours such as aggression, grooming, and disease‑related inactivity.