Terrain Type Detection for Smart Equine Gait Analysis Systems Using Inertial Sensors and Machine Learning
Authors: J. Parmentier, F. S. Bragança, Elin Hernlund, B. J. van der Zwaag
Journal: 2023 19th International Conference on Distributed Computing in Smart Systems and the Internet of Things (DCOSS-IoT)
Summary
# Editorial Summary Lameness evaluation in horses relies heavily on observing movement across different terrain types, yet current IMU-based gait analysis systems fail to automatically classify whether data is collected on hard or soft ground—a critical variable that directly influences the apparent severity of pain-related gait alterations. Researchers analysed inertial sensor data from 111 horses wearing multiple IMU units (placed at withers, pelvis, and all four limbs) using machine learning and deep learning classifiers, testing various feature-extraction approaches and sampling frequencies ranging from 10 to 200 Hz to assess real-world feasibility. Their Convolutional Neural Network models successfully identified terrain type with high accuracy using only a single accelerometer mounted on the front limb, and crucially, downsampled signals performed comparably to full-frequency data, indicating that real-time processing on portable devices is achievable without computational burden. For veterinary and rehabilitation professionals, this development means future smart gait analysis systems can automatically contextualise lameness observations by accounting for terrain effects, improving diagnostic consistency and enabling more reliable longitudinal monitoring of individual horses; farriers and physiotherapists can also expect clearer data about how their interventions influence movement across the varied surfaces horses encounter in everyday work. The pathway to integrating automated terrain detection into commercial gait monitoring platforms is now technically viable, potentially transforming how movement changes are interpreted and tracked throughout rehabilitation and training programmes.
Read the full abstract on the publisher's site
Practical Takeaways
- •Future wearable gait monitoring systems can automatically detect terrain type during lameness evaluation, eliminating information loss and improving diagnostic accuracy
- •Single-sensor systems reduce cost and complexity while maintaining performance, making deployment more practical in field settings
- •Real-time terrain detection capability enables better interpretation of gait changes across different surface types, supporting more objective lameness localization
Key Findings
- •Convolutional Neural Network models accurately classified terrain types (hard vs soft) using IMU data from 111 horses
- •A single IMU sensor placed on the front limb was sufficient for terrain classification, reducing hardware complexity
- •Downsampled signals (lower sampling frequencies) maintained classification accuracy, enabling real-time applications on resource-limited devices