AiCoachU – Artificial intelligence is coaching you
Physical activity is one of the key contributors to health and quality of life. Running is popular and an efficient and affordable modality of physical activity. However, if done improperly, it may induce injuries leading to lower life quality and additional health and social costs. Therefore, it is important to provide tools for effective and injury-free physical activity. In the present study, a new generation of IMU sensor patches with considerably smaller dimensions and weight were employed for rearfoot and pelvis stability measurement and their changes due to fatigue during running at different velocities and surface inclinations. This was analysed through the pelvis and rearfoot motion patterns employing deep learning.
A software platform for capturing, storing, synchronization and processing of the captured data was developed as an integral part of the project. Based on an STM32 SensorTile module with BLE wireless connectivity, the platform includes custom firmware built on a FreeRTOS kernel and an Android application for device management. A dedicated synchronization algorithm aligns data streams from multiple sensors to sub-millisecond accuracy. The platform was connected to the hardware acquisition system and enabled both automatic data access and access through a user interface.
Standardised laboratory treadmill protocols were carried out, systematically varying speed (10–14 km/h), inclination (0–10 %), and fatigue state. Kinematic data were captured using a gold-standard 3D motion capture system (Qualisys, 12 cameras), synchronised with the sensor patches. The collected dataset enabled a biomechanical analysis of rearfoot eversion/inversion as a potential injury-risk indicator for the ankle and Achilles tendon, as well as a 3D analysis of pelvis rotation, tilt, and frontal-plane drop as a key link in the lower-limb loading chain. Results demonstrate successful recognition of fatigue onset and excessive pelvic and rearfoot mechanics across different running conditions using deep learning, supporting the development of a virtual running coach for safe running and informed shoe selection. Part of the dataset has been published as open research data to support further validation and reuse by the wider research community.
Partners
Faculty of Electrical Engineering - University of Ljubljana (FE)
Faculty of Sport - University of Ljubljana (FSB)
Jozef Stefan Institute (JSI)
P-Lab team
Funding
