A hidden Markov model-based stride segmentation technique applied to equine inertial sensor trunk movement data.

The research article talks about the application of Hidden Markov Models (HMMs) in stride segmentation of horses. Through these models, researchers can identify steps taken from data collected by a trunk-mounted sensor on running horses for more efficient, automated data processing.

Study Background and Purpose

• As inertial sensors become smaller and lighter, they are now more commonly used for tracking movement data in both humans and animals. Despite advances in hardware, managing and processing large amounts of the data gathered from these sensors remains challenging.
• The study aims to address this processing challenge by applying Hidden Markov Models (HMMs), a stochastic pattern recognition method commonly used in processing and classifying non-stationary data. Specifically, the researchers utilized HMMs to identify strides in data collected from a trunk-mounted sensor on galloping Thoroughbred racehorses.
• The primary purpose of the study is to develop automated processing techniques to address the increasing reliance on inertial sensors for measuring locomotion in both clinical and non-clinical settings.

Methodology and Results

• Data was collected from seven Thoroughbred racehorses performing mixed gait sequences. The dataset was split into training, cross-validation and independent test subsets.
• Stride segmentations were manually created and used for training the HMMs, as well as for assessing cross-validation and test set performance.
• The results showed a high degree of accuracy, with 91% of test set strides accurately detected as lying within +/- 40 ms of the manually segmented stride starts. Furthermore, the automated system did not miss any strides.
• Despite the high accuracy, the system did overshoot by identifying extra strides at the beginning of each trial run.

Conclusions and Recommendations

• The use of HMMs can be seen as an effective tool for classifying and processing large amounts of locomotion data collected from inertial sensors.
• The notably high accuracy indicates that these automated processing techniques could enhance large-scale data processing efficiency and facilitate quicker decision-making.
• The researchers suggest broader use of HMM-based classifiers for stride segmentation, pointing out their straightforward implementation and robust performance with limited training data.