Potential of mobile phones for large-scale driving studies demonstrated

Data labeling with the system application TöltSense. The left panel shows the position of the sensors on the lower legs. The right panel shows the mobile phone sensor data after rotation to a specific frame of reference (world frame) and the gait designations when a horse changes from walk to tölt. The X and Y curves correspond to variations in the acceleration and gyroscope signals in the horizontal plane, and the Z curve corresponds to variations in the vertical axis. The sign of the signal on a given axis corresponds to the direction of the signal on that axis. The highlighted segment shows an example of the input used for the machine learning model. Image: Davíðsson et al. https://doi.org/10.3390/ani13010183

Equestrian activities can be studied extensively using cellphones to collect data on gaits, researchers report.

The researchers reported the results of their study in which they examined the use of cellphone sensors to accurately classify the gaits of five-gaited horses.

Haraldur Davíðsson and his colleagues write in the magazine Animals, said that mobile devices have become an accepted part of life. With rapid technological advances, their applications are constantly evolving.

Many phones have sophisticated built-in motion sensors

Automated gait classification has traditionally been studied using horse-mounted sensors, they said, but modern cellphones now have the potential for such use. In fact, gait classification has already been implemented in commercial smartphone apps.

The researchers said that while smartphone-based sensors are more accessible, the performance of gait classification models using data from such sensors is not widely known or accessible.

Davíðsson, Torben Rees, Marta Rut Ólafsdóttir and Hafsteinn Einarsson set out to perform horse gait classification using deep learning models and data from sensors in a mobile phone carried in the rider’s pocket.

Machine learning has enabled data from mobile phones carried in the pocket of drivers to be classified by gait.
The team focused their gait study on the Icelandic horse. Photo by dalli58

They wanted to determine the accuracy of gait classification for all five Icelandic horse gaits using models trained on data from the accelerometer and gyroscope in mobile phones that the rider carries in their pockets.

The study team focused on the Icelandic horse, which, in addition to the three standard gaits – walk, trot and canter – can execute two additional gaits, tölt and pace.

Information for the study was collected from a horse farm in southern England and from various horse farms and training centers in Iceland.

17 Icelandic horses and 14 riders were used for the measurements, and the phone was placed in a pocket on the rider’s clothing, chosen by the rider. The location of the phone varied between drivers, with the phone tucked into either pants or jacket pockets.

The horses were ridden on a variety of outdoor surfaces, on a track, a clay court, or a trail.

The information was collected simultaneously from the riders’ mobile phones – either Samsungs or iPhones – and a commercial gait classification system based on four wearable sensors attached to the horse’s limbs. The TöltSense system used by the researchers is a training tool designed to classify and analyze the quality of Icelandic horses’ gaits and provide real-time feedback to the rider.

The authors said that using the commercial gait classification system proved to be a cost-effective way to collect gait designations alongside cellphone data without external help or environmental restrictions.

In this way, 5.8 hours of time-coordinated and gait-tagged data were collected, corresponding to thousands of short segments for each gait.

A machine learning model was then trained to classify the gaits from the phone’s accelerometer and gyroscope, achieving 94.4% accuracy in classifying the Icelandic horse’s five gaits.

The authors said the most common confusion in the model is between tölt and trot.

“It is conceivable that more training data from different horses and riders would improve performance at the trot, as performance on the training set was better than on the test set.”

In a separate study, the researchers showed a very high level of gait agreement between the TöltSense system and four qualified sports judges who classified Icelandic horses’ gaits in video segments.

Using cut-offs in transitions can push approval above 99%, but approval was about 94% even without cut-offs, they said.

Overall, the results suggest that equestrian activities can be studied on a large scale using cell phones to collect data on gaits, the study team concluded.

“Although our study showed that mobile phone sensors could be effective for gait classification, there are still some limitations that need to be addressed in future research.

“For example, further studies could examine the effects of different driving styles or equipment on the accuracy of gear classification, or explore ways to minimize the impact of factors such as phone placement.

“By addressing these questions, we can further improve our understanding of horse gait and its role in riding.”

The research, they said, shows cell phones could help reduce the cost of large-scale gait studies.

“This efficient method of capturing tagged data will be invaluable for ongoing research into equestrian activities.”

Davíðsson is at both the University of Iceland’s Department of Computer Science and at TöltSense Ltd. Rees is at Horseday ehf in Reykjavik, Iceland; Ólafsdóttir is at TöltSense Ltd; and Einarsson is at the Faculty of Computer Science at the University of Iceland.

Davíðsson, HB; Rees, T.; Ólafsdóttir, MR; Einarsson, H. Efficient development of gait classification models for horses with five gaits based on mobile phone sensors. Animals 2023, 13, 183. https://doi.org/10.3390/ani13010183

The study, published under a Creative Commons Licensecan be read here.