The adoption of artificial intelligence is not limited to natively digital industries or highly automated sectors like manufacturing. Even more traditional fields, such as agriculture and livestock farming, are now undergoing a deep transformation, driven by data and automation.

In Switzerland, where livestock farming is a long-standing and strategic part of the economy, Artificialy has developed an innovative Computer Vision solution to improve herd management and optimize production efficiency.

Automating Animal Tracking and Sorting

The project originated from the need to automate the gathering and management of dairy cows. In modern farms, cows are milked twice a day, and some require specific treatments for health or reproductive reasons. Identifying and separating these animals manually is labor-intensive, costly, and time-consuming.

Traditional infrared beam systems could only detect the presence of cows and activate a sorting gate, but this approach had clear limitations: if an animal stopped in the wrong position or stood too close to another, the system could easily malfunction or produce errors.

From this challenge, we developed an intelligent, real-time AI tracking system capable of operating reliably even in complex environmental conditions.

Our AI Solution: Computer Vision and Predictive Control

We designed a deep learning–based Computer Vision system capable of recognizing and tracking each animal individually.

The system uses a 360° fisheye camera with IR illumination, integrated with an NVIDIA Jetson Orin computing platform to ensure local processing and immediate response times.

The AI architecture relies on three main components:

1. Object Detection – Identifies and locates each cow within the field of view.

2. Landmarks Detection – Recognizes key body parts (head, shoulders, tail) to enhance tracking accuracy.

3. Temporal Tracking and Predictive Control – By combining visual data over time (T, T+1, T+2), the system predicts movements and activates gates proactively, minimizing errors and waiting times.

The result is a continuous, automated flow in which animals are identified, tracked, and guided along the correct path safely and efficiently. The system can even re-identify animals temporarily hidden from view.

The Challenges of Computer Vision in Real Environments

A barn is a highly dynamic environment: cows differ in appearance, lighting changes throughout the day, cameras can easily get dirty, and animals behave unpredictably.

Overcoming these challenges was essential to ensure a robust, reliable, and field-ready system.

An especially interesting aspect of this project was the use of Generative AI to increase the robustness of the Computer Vision model. Since the camera was fixed in a single position and environmental context, its performance could degrade in different scenarios. To address this, we used image inpainting techniques based on Stable Diffusion, generating synthetic variations of the scene—grass fields, asphalt, and different soil types. This approach expanded and diversified the dataset, improving the model’s ability to generalize across varied real-world conditions.

Results: Efficiency, Animal Welfare, and Sustainability

Thanks to this solution, the entire drafting process became more efficient, precise, and animal-friendly. Automated gate control reduced both the number of incidents and the stress levels of the cows, directly contributing to better herd well-being and higher productivity.

Moreover, the system helped optimize workflows and cut the energy footprint associated with manual operations—a concrete example of sustainable innovation applied to the agricultural sector.

AI as a Driver of Swiss Agritech Innovation

This project demonstrates how artificial intelligence can deliver real value even in traditional industries. The data-driven and engineering-driven approach of Artificialy bridges the gap between AI research and production, making advanced technologies accessible and scalable in both industrial and agricultural contexts.

In the broader agritech landscape, innovations like this pave the way toward a new era of smart farming, where every decision, from livestock movement to equipment maintenance, can be optimized through data.