At an ICAR farm in Nagaland, cameras and artificial intelligence are learning to read the everyday behaviour of Mithun. The technology could eventually help farmers with breeding, animal health and round-the-clock monitoring.
Nagaland: In the quiet hours of a Mithun (Bos frontalis) shed, an animal may be feeding, standing, lying down or mounting another animal. For a farmer watching a herd, these are ordinary movements. But they can also carry important clues about an animal’s health, comfort and reproductive condition.
Now, researchers are teaching a computer to notice those clues.
At the ICAR-National Research Centre on Mithun (NRC-Mithun) farm in Nagaland, researchers have developed an artificial intelligence-based system that can detect and track the behaviour of Mithun in real time. The study, published recently in Engineering Research Express, describes the work as the first AI-based, real-time, non-contact framework developed for automatic Mithun behaviour detection and tracking in a natural farm environment. For communities across Northeast India, this is more than a technology story.
Mithun, known as the “Cattle of the Hills”, has long been woven into the social and economic life of tribal communities. It is associated with wealth, barter, dowry, rituals and community feasts, while also providing meat and supporting livelihoods.
Yet keeping a close watch on every animal is difficult. Traditional behaviour monitoring depends heavily on people observing animals manually. That takes time and labour, and it becomes particularly challenging when animals need to be monitored continuously, including at night.
This is where ICAR’s farm has become an important part of the experiment. The researchers used footage from 12 high-definition CCTV cameras installed across two sheds at the NRC-Mithun farm. The cameras operate continuously, providing day and night coverage, including infrared footage. From this surveillance system, researchers created a dataset of 3,000 manually annotated images showing four key behaviours: feeding, standing, lying and mounting.
The AI system has two main jobs. The first, powered by YOLOv8n, identifies what the animal is doing. The second, using DeepSORT, follows individual animals from one frame to another and assigns them persistent identities. Put simply, the computer is being taught to answer two questions: “What is this Mithun doing?” and “Which Mithun is this?”
The results were striking. The YOLOv8n model recorded a 99.5% mean average precision at the standard mAP@0.5 measure, with 99.6% recall. Detection ran at about 31 frames per second on an NVIDIA RTX 3060, allowing the system to operate at real-time speed.
The researchers also tested the system in difficult conditions, including partial occlusion, background clutter, uneven or wet ground, shadows, motion blur and nighttime infrared footage.
Why does that matter to a farmer? Because behaviour can tell a story before an animal visibly appears sick.
Feeding, standing and lying patterns can provide clues about comfort, nutrition and physiological condition. Mounting behaviour, meanwhile, is particularly important because it can provide useful information for reproductive and oestrus management.
If such systems can eventually be deployed reliably on farms, farmers and livestock managers could have access to continuous behavioural information without having to physically observe animals throughout the day and night. For remote and resource-constrained livestock systems, that could be significant.
But the technology is not yet a ready-made solution for farmers. The researchers caution that the system has so far been tested on a single farm. Its performance still needs to be established across different farms, regions, seasons, stocking densities and camera arrangements. Severe occlusion can also affect detection and tracking. Moreover, the study covers only four behaviours, and tracking has not been quantitatively evaluated using standard identity-tracking measures because identity-level ground truth was unavailable.
That makes the role of ICAR particularly important. The NRC-Mithun farm provided something an AI laboratory cannot easily recreate: a real Mithun environment, real animals and continuous farm surveillance. The research therefore brings together livestock science and computer vision in a setting directly connected to the animal and the communities that depend on it. The researchers now envisage expanding the system to more behaviours, including aggression, grooming and disease-related inactivity, while exploring temporal AI models, edge-device deployment and larger datasets from different farms and seasons.
The promise is not that a machine will replace the farmer. It is that a machine could become another pair of eyes. For a farmer whose herd may need watching long after daylight fades, that could mean something as simple and valuable as knowing that an animal has stopped feeding, changed its normal activity or displayed reproductive behaviour.
In the hills where Mithun has been part of community life for generations, the next generation of animal care may therefore involve an unlikely new farmhand: artificial intelligence, watching quietly from a camera, 24 hours a day. The camera may see the Mithun. But the larger goal is to help the farmer understand what the animal is saying.
(Source: ICAR-NRC on Mithun, Nagaland)
Like on Facebook
Subscribe on Youtube
Follow on X X
Like on instagram