Machine Learning for Running Analysis with Wearable Inertial Sensors

Ground contact time and stride length analysis is crucial in high performance running. However, the equipment currently available in the market to perform such analysis, like optical sensors, is often only accessible to professional athletes due to their high cost and the complexity of installation. This study aims to explore a cost-effective alternative to optical sensors by utilizing inexpensive inertial measurement units placed on the athlete's ankles, offering a wearable solution accessible to a larger audience. A machine learning approach was used to address the problem, with the creation of a training set that associates accelerometer and gyroscope recordings when the athlete is running, to external measurements of ground contact time, obtained with pressure sensors in the insoles, and stride length, evaluated with a videocamera. The training set has been used to create a model that will predict these values solely based on the data from the inertial measurement units. The results demonstrate the suitability of this type of data for a machine learning environment and, with the proper model training and training set, it is possible to achieve results comparable to those obtained using optical sensors.