AI-Based Detection of Behavioral and Condition Indicators in Horses in Stables

Initial Situation and Objectives

Horses—especially those kept in individual stalls—often spend long periods of time without supervision. Important indicators of stress, discomfort, or acute health problems frequently go unnoticed or are detected too late. Continuous and objective observation is difficult to implement under real-world conditions, even though it would be highly valuable for animal safety and early intervention.

The aim of the project is to develop a non-invasive, camera-based system for the automated detection of behaviors and specific conditions of horses in stalls. The focus is on the continuous monitoring of individual animals in order to identify both acute situations and long-term behavioral abnormalities at an early stage.

The work is based on support provided through the Fraunhofer internal AHEAD Deeptech Accelerator, which helps transform ideas and associated technologies into viable business models.

Technological Approach

The project uses modern computer vision and AI methods for the automated analysis of video data recorded in horse stalls. Key components include:

• object detection and classification of specific states and activities of horses
• training models for on-edge computing to enable reliable analysis directly on-site
• targeted data enhancement and feature design optimized for evaluating temporal behavior patterns using time-series analysis

The models are developed under real housing conditions and validated together with domain experts. The system aims to detect, among other things, general stress as well as potentially critical conditions such as symptoms of colic.

Benefits, Application, and Exploitation

The solution enables 24/7 monitoring of horses in stalls and supports both the detection of acute abnormalities and the analysis of long-term behavioral changes. This allows horse owners to react at an early stage and gradually implement measures to improve housing conditions and animal welfare.

The developed models and data structures are specifically designed for a B2C application for private horse owners. In the long term, this will lead to a product for monitoring the condition and activity of horses in stalls.

At the same time, the project serves as an MVP foundation for a potential spin-off in the field of digital animal welfare monitoring.

Practical Environment

Development and validation are carried out under real-world conditions in collaboration with the Redefin State Stud (Landgestüt Redefin).

Funding: