Researchers at the ICAR-National Research Centre on Mithun (ICAR-NRC on Mithun), Nagaland, have developed an artificial intelligence-based system that can detect and track the behaviour of Mithun in real time, opening up possibilities for improved animal health, welfare and breeding management.
The system uses cameras and artificial intelligence to automatically identify four key behaviours of Mithun – feeding, standing, lying and mounting – while simultaneously tracking individual animals.
Mithun, popularly known as the “Cattle of the Hills”, has considerable social, cultural and economic importance among tribal communities across Northeast India and contributes to livelihoods and food security in the region.
The research team deployed 12 high-definition CCTV cameras across two sheds at the ICAR-NRC on Mithun farm in Nagaland, providing continuous day-and-night surveillance, including infrared coverage. From the recorded footage, the researchers developed a dataset of 3,000 manually annotated images covering the four behaviours.
The AI system combines the YOLOv8n model for behaviour detection with DeepSORT technology to track individual animals and maintain their identities across video frames.
According to the researchers, the detection model achieved a mean average precision of 99.5 per cent at mAP@0.5, with a recall of 99.6 per cent. The system processed footage at approximately 31 frames per second using an NVIDIA RTX 3060 graphics processing unit, demonstrating its potential for real-time monitoring.
The system was also tested under challenging farm conditions, including partial obstruction of animals, background clutter, uneven and wet ground, shadows, motion blur and nighttime infrared footage.
The researchers said continuous monitoring of behaviour could provide useful information on the health and physiological condition of animals. Changes in feeding, standing and lying patterns may indicate changes in health, comfort or nutrition, while mounting behaviour could provide inputs for reproductive and oestrus management.
Such automated monitoring could reduce the need for constant manual observation, particularly during night hours, and provide livestock managers with behavioural information on animals over extended periods.
However, the researchers noted that the system has so far been evaluated at a single farm and requires further validation across different farms, geographical regions, seasons, stocking densities and camera arrangements. Severe obstruction of animals can also affect detection and tracking performance.
The team plans to expand the system to identify additional behaviours, including aggression, grooming and disease-related inactivity. Future research may also examine temporal AI models, edge-device deployment and larger datasets covering different farms and seasons.
The study was conducted by researchers from ICAR-NRC on Mithun, Nagaland, in collaboration with NIT Nagaland, Nagaland University and CHRIST (Deemed to be University).
The research has been published in Engineering Research Express, Volume 8 (2026), Article 175213.