Using Different Combinations of Body-Mounted IMU Sensors to Estimate Speed of Horses-A Machine Learning Approach

Area of Science:

• Biomechanics
• Locomotion research
• Machine learning applications

Background:

• Accurate speed estimation is crucial for biomechanical analysis and locomotion research.
• Existing methods like GPS and inertial measurement units (IMUs) have limitations, including signal dependency and accumulated errors.
• Developing robust speed estimation techniques for equines is essential for research and performance analysis.

Purpose of the Study:

• To develop and evaluate machine learning (ML) models for estimating horse speed using body-mounted IMUs.
• To assess the impact of IMU location on speed estimation accuracy across different breeds and gaits.
• To compare the performance of various ML algorithms for equine speed estimation.

Main Methods:

• Trained ML models using data from 40 Icelandic and Franches-Montagnes horses across five gaits (walk, trot, tölt, pace, canter).
• Utilized signals from seven body-mounted IMUs placed on the sacrum, withers, head, and limbs.
• Evaluated model accuracy (RMSE) per gait and compared ML algorithms and IMU placement strategies.

Main Results:

• Achieved high accuracy in horse speed estimation with a root mean square error (RMSE) of 0.25 m/s.
• Demonstrated that ML models provided accurate estimations regardless of IMU location on the body.
• The developed models showed superior accuracy compared to most existing equine and human speed estimation literature.

Conclusions:

• Highly accurate horse speed estimation models were successfully developed using ML and IMU data.
• The models are independent of specific IMU locations on the body and gait type.
• This approach offers a reliable and practical alternative for speed measurement in equine biomechanics and locomotion research.