Researchers at ICAR-National Research Centre on Mithun (ICAR-NRC on Mithun) in Nagaland have developed an artificial intelligence-based system for real-time detection and tracking of Mithun (Bos frontalis) behavior in natural farm environments. The research, published in Engineering Research Express, presents an AI-based, non-contact framework for automatic monitoring of the animal known as 'Cattle of the Hills', which holds significant social, cultural and economic importance for tribal communities across Northeast India.
The system utilizes 12 high-definition CCTV cameras deployed across two sheds at the ICAR-NRC on Mithun farm, providing continuous day-and-night surveillance including infrared coverage. The research team developed a dataset comprising 3,000 manually annotated images representing four key Mithun behaviors: feeding, standing, lying, and mounting. The AI framework combines the YOLOv8n model for behavior detection with DeepSORT technology for tracking individual animals across video frames and assigning persistent identities.
Technical performance metrics show the YOLOv8n model achieved a mean average precision of 99.5% at mAP@0.5, with a recall of 99.6%. The system processes approximately 31 frames per second on an NVIDIA RTX 3060 graphics processing unit, demonstrating real-time application capability. The framework was tested under challenging farm conditions including partial occlusion, background clutter, uneven and wet ground, shadows, motion blur, and nighttime infrared footage, with specific examples showing 96% confidence for detecting lying and feeding behaviors simultaneously and 97% confidence for mounting behavior in infrared.
The technology has significant potential for improving livestock monitoring and management, as changes in feeding, standing, and lying patterns can provide early indications of animal health, comfort, nutrition and physiological condition, while mounting behavior supports reproductive and oestrus management. Continuous automated monitoring enables farmers and livestock managers to access behavioral information without requiring constant physical observation throughout day and night.
Researchers noted the system requires further validation across different farms, geographical regions, seasons, stocking densities and camera arrangements. The current study focuses on only four behaviors, and severe occlusion can affect detection and tracking performance. Future work may expand the AI framework to identify additional behaviors such as aggression, grooming and disease-related inactivity, and explore temporal AI models, edge-device deployment, and larger datasets covering different farms and seasons.
The research was conducted by ICAR-NRC on Mithun, Nagaland, in collaboration with NIT Nagaland, Nagaland University and CHRIST (Deemed to be University), and published in Engineering Research Express, Volume 8 (2026), Article 175213.
This content may not be financially relevant.