Image-Based Health Monitoring in Dairy Cattle Barns

Problem and Objective

The early detection of health impairments is a key prerequisite for both animal welfare and economic efficiency in dairy farming. In larger herds, however, the assessment of animals’ health status in everyday barn operations is still largely carried out manually and only when a specific issue becomes apparent. This process is labor-intensive, highly dependent on experience, and often begins only once visible changes have already become pronounced. Consequently, there is a clear need for objective, continuously deployable, and scalable monitoring methods.

Against this background, a camera-based, non-invasive health monitoring system is being developed that systematically captures movement, vital, and behavioral data and integrates them into a robust decision-support basis for herd management. Development and testing are conducted under real barn conditions in three dairy farms with a total of around 4,000 cows, including close collaboration with the Research Institute for Farm Animal Biology in Dummerstorf.

Automated Lameness Detection

An initial application-focused priority is automated lameness detection. RGB cameras are installed in highly frequented areas of the barn and continuously record the gait patterns of all animals in the herd. Computer vision methods identify relevant body regions and derive parameters such as angles, speed, and acceleration.

The model is designed to operate robustly under varying lighting conditions, different barn environments, and individual animal differences. Lameness detection therefore represents not an isolated use case, but rather a functional entry point into a broader health monitoring system.

Non-Invasive Measurement of Vital Parameters

Based on the same technological foundation, additional physiological indicators are integrated. Thermal cameras enable contactless temperature measurement, while RGB-based methods support image-based, non-invasive pulse analysis using imaging photoplethysmography (iPPG).

In addition, an image-based respiratory rate detection system is being developed, which Fraunhofer IGD is advancing as an associated partner together with consortium partners in the KI-TIERWOHL project. Movement and vital parameters are thus captured within a unified system context.

Pose Detection and Behavioral Analysis

At the same time, continuous pose detection is being developed as the basis for behavioral analysis. Camera-based methods detect postures such as lying, standing, feeding, or drinking. Time-series analyses evaluate the duration, frequency, and transitions between poses and identify deviations from an animal’s individual baseline behavior.

Changes in activity and resting behavior can therefore be detected at an early stage and correlated with physiological abnormalities.

Externally Visible Condition Indicators

In addition, externally visible characteristics such as dirt contamination and injuries are automatically detected. Image-based methods evaluate their extent and distribution and thus provide additional indications of hygiene status and housing conditions as complementary information for health management.

Individual Animal Identification as an Integration Key

The integration of these information streams requires the clear identification of individual animals using purely visual methods. For this purpose, a biometric facial recognition system for dairy cows is being developed, enabling all recorded movement, vital, and behavioral data to be assigned to specific animals.

On this basis, a continuous digital health record for each animal in the barn environment is gradually established.

Transferability and System Integration

The development is carried out using the dairy barn context as a representative application scenario. However, the underlying methodological principles—camera-based condition monitoring, multimodal data fusion, and individual animal identification—are designed to be transferable to other animal species and husbandry systems.

The overall system follows a modular design and can function either as a standalone application or as an integrable component for existing herd management solutions. It is particularly aimed at larger herds with high requirements for documentation, transparency, and scalability, and is designed with future transfer to other animal species and housing systems in mind.