New Delhi: Researchers at ICAR-NRC on Mithun in Nagaland developed an AI Mithun tracking system for real-time animal behaviour monitoring.
The system used cameras and artificial intelligence to detect and track Mithun behaviour in a natural farm setting. Researchers said it could support animal health, welfare and breeding.
The study appeared in Engineering Research Express. It presented a non-contact system for automatic detection and tracking of Mithun (Bos frontalis).
Mithun, known as the “Cattle of the Hills”, has major social, cultural and economic value in Northeast India. The animal also supports livelihoods and food security among tribal communities.
Researchers noted that behaviour changes can offer early clues about animal health and comfort. Feeding, standing, lying and reproductive activity can also reflect nutrition and physiological condition.
However, manual observation requires significant labour. It also becomes difficult to monitor animals continuously, especially at night.
To address that challenge, the team installed 12 high-definition CCTV cameras across two sheds at the ICAR-NRC on Mithun farm. The cameras provided continuous day-and-night coverage, including infrared surveillance.
The researchers created a dataset of 3,000 manually annotated images from the recorded footage. Those images covered four behaviours: feeding, standing, lying and mounting.
The AI Mithun tracking framework combined the YOLOv8n model with DeepSORT technology. YOLOv8n detected behaviour, while DeepSORT tracked individual animals across video frames.
The system assigned persistent identities to individual Mithun. It could therefore identify behaviour while tracking each animal in real time.
Testing showed strong detection results. YOLOv8n achieved 99.5 per cent mean average precision at mAP@0.5. Its recall reached 99.6 per cent.
The system processed about 31 frames per second on an NVIDIA RTX 3060 graphics processing unit. This result demonstrated its potential for real-time farm use.
AI Mithun tracking expands livestock monitoring
Researchers also tested the system under difficult farm conditions. These included partial obstruction, background clutter, uneven and wet ground, shadows and motion blur.
The system also handled nighttime infrared footage during testing. Such conditions can make continuous animal monitoring more difficult.
The technology could help farmers and livestock managers monitor behaviour without constant physical observation. Feeding, standing and lying patterns could provide useful health and welfare signals.
Mounting behaviour could also support reproductive and oestrus management. Continuous monitoring could provide behavioural data throughout the day and night.
However, the researchers said the system still needs wider validation. The study tested it at only one farm.
Future trials must cover different farms, regions, seasons, stocking densities and camera arrangements. Severe obstruction can also reduce detection and tracking performance.
The current study covers only four behaviours. Researchers said they also need to assess identity tracking quantitatively using standard tracking measures.
Future work could expand the AI Mithun tracking system to detect aggression, grooming and disease-related inactivity. Researchers may also test temporal AI models and edge-device deployment.
They also plan to build larger datasets covering different farms and seasons. These improvements could strengthen automated livestock monitoring.
The study showed how artificial intelligence and computer vision could support precision livestock farming. Such systems could give farmers continuous, data-based information about animal behaviour.
The research appeared in Engineering Research Express, Volume 8 (2026), Article 175213. Researchers from ICAR-NRC on Mithun conducted the study with NIT Nagaland, Nagaland University and CHRIST (Deemed to be University).
