Development and Validation of an Automated Video Tracking Model for Stabled Horses.
Authors: Kil Nuray, Ertelt Katrin, Auer Ulrike
Journal: Animals : an open access journal from MDPI
Summary
# Editorial Summary: Automated Video Tracking for Stabled Horses Behavioural changes represent one of the most reliable indicators of pain and compromised welfare in horses, yet systematic assessment remains challenging in clinical settings. Kil, Ertelt and Auer developed and validated an automated video tracking system using convolutional neural networks (specifically the Loopy model) to detect key anatomical landmarks from time-lapse footage recorded in hospital box stalls, addressing the need for objective, continuous monitoring without human observation bias. The system successfully identified three critical keypoints—nose, withers and tail—with sensitivity exceeding 80% and acceptable error margins of 2–7% depending on the landmark, demonstrating sufficient accuracy for reliable kinematic analysis. These validated tracking outputs enable quantitative assessment of movement patterns, postural changes and locomotor abnormalities that might otherwise be missed during brief clinical examinations, directly supporting earlier pain recognition and objective quality-of-life monitoring. For equine practitioners, this technology offers the potential to streamline behavioural assessment protocols, generate longitudinal data on individual horses' activity and posture, and ultimately support the development of automated algorithms for detecting pain-related behavioural signatures in both clinical and competition settings.
Read the full abstract on PubMed
Practical Takeaways
- •Automated video tracking systems could provide objective, continuous monitoring of stabled horses to detect pain-related behaviour changes without observer bias
- •The high sensitivity (>80%) and low error rates (2-7%) suggest this technology is sufficiently accurate for practical clinical application in equine hospitals and facilities
- •This technology could enable early recognition of pain conditions by identifying subtle behavioural deviations from normal patterns, improving welfare and treatment outcomes
Key Findings
- •Convolutional neural network (Loopy) successfully detected key body points (nose, withers, tail) with >80% sensitivity
- •Error rates for automated detection ranged from 2-7% depending on the key point analyzed
- •Automated video tracking from action camera footage in time-lapse mode enables detection of movement patterns in box stalls
- •Automated behaviour detection model has potential to improve pain recognition and quality of life assessment in stabled horses