Assessment of Real-World Running With Lab-Validated Algorithms Based on Foot-Worn Inertial Sensors

Following the cultural revolution of the late 1960s, the number of elite and recreational runners rose consistently, reaching approximately 7.9 million road races participants in 2018. Today, running is everywhere. City parks, forests, mountain trails, and athletic tracks are now the playground of numerous running enthusiasts, whatever their ages, gender, and social background. With such heterogeneity in the runners' profiles, the motives to maintain a running habit vary from psychological, social, and physical objectives to performance-oriented goals. Although the health benefits of running are well-recognized in the scientific literature, its regular practice also presents risks of injuries. To study the underlying mechanisms associated with injuries or improvements in performances, scientists have investigated the kinematics and the kinetics characteristics of the running gait. Habitually, this quest requires the use of precise monitoring instruments only accessible in well-equipped research laboratories. However, over the past two decades, the advent of wearable sensors shifted the analysis of running into real-world settings, where runners encounter different environments, outside, in the wild. It is in this setting that the current thesis situates itself. This thesis presents a new wearable system for the objective assessment of the running gait in real-world conditions. The proposed method uses foot-worn inertial sensors and lab-validated algorithms to provide a reliable analysis of the spatiotemporal parameters of running. The system can operate outdoors and over extended periods while providing a quasi-real-time evaluation of each step. Further, with its automatic detection of the sensor location and calibration, the proposed method is easy-to-use and accessible to non-initiated users. For the technical validation of the proposed system, the spatiotemporal metrics were compared with gold-standard reference systems. Temporal events and gait phases were validated in-lab against a reference force plate integrated into a treadmill, and the results of a novel orientation-drift correction model compared to a state-of-the-art motion capture system. Three overground speed estimation methods were evaluated in real-world conditions and compared to a Global Navigation Satellite System device. Finally, these methods were tested in different settings, such as a marathon race, a mountain ultra-marathon, and a 400-m hurdling competition. These tests provided valuable insight into the limitations of the proposed system and suggested several improvements for its use in real-world conditions. Overall, this thesis aims to augment the resolution of running analysis by handling the technical challenges associated with inertial sensors and providing fast and reliable biomechanical metrics. As such, the system could contribute to extending the knowledge about the mechanical adaptations experienced in real-world environments and long-term running. Moreover, the