# Detecting Equine Gaits Through Rider-Worn Accelerometers

## TL;DR

This study uses accelerometers on riders to detect horse gaits, avoiding direct sensor attachment to horses.

## Contribution

The study introduces a novel rider-centric approach using a ConvLSTM2D model for accurate horse gait classification.

## Key findings

- A ConvLSTM2D model achieved 89.72% accuracy in classifying four horse gaits.
- Optimal performance was observed with a four-second interval and 25 Hz sampling frequency.
- The model achieved an F1-score of 86.18% using LOSOCV validation.

## Abstract

Understanding how horses move can improve training practices and support equestrian health, but attaching sensors directly to them may cause discomfort and interfere with natural movement. To overcome this, our study placed sensors on riders instead of horses to classify different horse gaits (halt, walk, trot, and canter). We tested four different sensor placements on the riders—the knee, backbone, chest, and arm—examining five riders and seven horses. Our research also explored how sensor settings, such as data collection speed and analysis interval, affect classification accuracy. After comparing eight different classification models, we found that a specialized neural network model performed best, correctly identifying horse movements with 89.7% accuracy. These findings help show how wearable technology can assist in monitoring horse movement accurately and comfortably, potentially benefiting horse training and welfare.

Automatic horse gait classification offers insights into training intensity, but direct sensor attachment to horses raises concerns about discomfort, behavioral disruption, and entanglement risks. To address this, our study leverages rider-centric accelerometers for movement classification. The position of a sensor, sampling frequency, and window size of segmented signal data have a major impact on classification accuracy in activity recognition. Yet, there are no studies that have evaluated the effect of all these factors simultaneously using accelerometer data from four distinct rider locations (the knee, backbone, chest, and arm) across five riders and seven horses performing three gaits. A total of eight models were compared, and an LSTM-convolutional network (ConvLSTM2D) achieved the highest accuracy, with an average accuracy of 89.72% considering four movements (halt, walk, trot, and canter). The model performed best with an interval width of four seconds and a sampling frequency of 25 Hz. Additionally, an F1-score of 86.18% was achieved and validated using LOSOCV (Leave One Subject Out Cross-Validation).