Horse Jumping and Dressage Training Activity Detection Using Accelerometer Data.
Authors: Eerdekens Anniek, Deruyck Margot, Fontaine Jaron, Damiaans Bert, Martens Luc, De Poorter Eli, Govaere Jan, Plets David, Joseph Wout
Journal: Animals : an open access journal from MDPI
Summary
# Editorial Summary: Activity Detection in Equine Training Using Accelerometer Technology Researchers from Belgium developed machine-learning models to automatically classify jumping and dressage training activities using leg-mounted accelerometer data collected from 14 well-trained horses, addressing a significant gap in current equine monitoring technology that struggles to distinguish between specific training movements rather than basic gaits. The neural network approach achieved near-perfect classification accuracy for jumping sessions (100%) and high accuracy for dressage activities (96.29%), with even greater precision when grouping related movements into superclasses—reaching 100% accuracy when consolidating 25 dressage movements into four broader categories. Beyond activity recognition, the model successfully identified the direction of lateral movements with 97.08% accuracy and estimated velocity during walk, trot and canter with root mean square errors of just 0.07, 0.14 and 0.42 m/s respectively. For practitioners, these findings suggest accelerometer-based systems could become practical tools for objectively monitoring training intensity, consistency and movement quality without requiring video analysis or subjective assessment, potentially enabling more data-driven approaches to performance tracking, fitness progression and early detection of asymmetries or compensatory patterns. Further development of user-friendly wearable devices based on these models could help farriers refine trimming and shoeing decisions, guide veterinary rehabilitation protocols, and support coaches in quantifying training load and detecting subtle changes in movement mechanics before they manifest as lameness or behavioural issues.
Read the full abstract on PubMed
Practical Takeaways
- •Wearable accelerometer technology can now reliably detect and classify specific jumping and advanced dressage movements beyond basic gaits, enabling objective performance tracking during training
- •Real-time velocity estimation from accelerometer data could provide riders and trainers with immediate feedback on gait quality and consistency without requiring expensive video analysis
- •This technology offers a user-friendly solution for monitoring training intensity and movement laterality, potentially supporting evidence-based training adjustments and fitness management
Key Findings
- •Jumping training activities classified with 100% accuracy using accelerometer-based neural network models
- •Dressage training activities classified with 96.29% accuracy across 25 different movements, improving to 100% when grouped into 4 superclasses
- •Direction of dressage movement identified with 97.08% accuracy using leg accelerometer data
- •Velocity estimation models achieved low root mean square errors of 0.07 m/s (walk), 0.14 m/s (trot), and 0.42 m/s (canter)