Efficient Development of Gait Classification Models for Five-Gaited Horses Based on Mobile Phone Sensors.
Authors: Davíðsson Haraldur B, Rees Torben, Ólafsdóttir Marta Rut, Einarsson Hafsteinn
Journal: Animals : an open access journal from MDPI
Summary
# Editorial Summary: Smartphone-Based Gait Classification for Icelandic Horses Researchers from Iceland developed a practical alternative to traditional wearable sensor systems by training a deep learning model to classify horse gaits using accelerometer and gyroscope data from a standard smartphone placed in the rider's pocket. Working with 17 horses and 14 riders, the team simultaneously collected data from the mobile phone and a reference system of four limb-mounted sensors, allowing them to create accurately labelled training data for their Bi-LSTM neural network model. The resulting classifier achieved 94.4% accuracy at identifying all five gaits of the Icelandic horse using only a 50 Hz signal from the phone's sensors, after rotating the data into the horse's frame of reference. This approach democratises gait analysis by eliminating the need for specialised equipment, opening possibilities for large-scale longitudinal studies of ridden horses and potentially enabling riders to monitor their own horses' movement patterns. Whilst promising, the authors note that future work should investigate how riding style, equipment variations and phone placement affect classification performance, as these practical variables will determine the real-world applicability of the technology for farriers, veterinarians and coaches seeking objective gait assessment tools.
Read the full abstract on PubMed
Practical Takeaways
- •Gait classification using readily accessible smartphone technology could enable large-scale data collection on riding activities and horse performance without specialized equipment
- •Smartphone-based gait monitoring may offer a cost-effective alternative to traditional wearable sensor systems for assessing horse movement quality during training and competition
- •Future work should address practical limitations including phone placement variability and effects of different riding styles before implementing this technology in field conditions
Key Findings
- •Mobile phone accelerometer and gyroscope sensors can classify five Icelandic horse gaits with up to 94.4% accuracy using a Bi-LSTM deep learning model
- •Data acquisition was efficient when combining smartphone sensors in rider's pocket with wearable sensors on horse limbs for generating labelled training data
- •A 50 Hz signal from phone sensors rotated to horse frame of reference provided sufficient input for accurate gait classification without additional features