As one of the three most popular sports in the Summer Olympics, competitive swimming has always been an attractive subject of study for sports scientists. The intricate nature of the swimmer's movements and the variety of techniques have led coaches to require analysis systems to gain a more detailed understanding of swimmers' performances, to plan training sessions as efficiently as possible and closely monitor the swimmer's progress. Surveys have shown that despite recent technological advances in sports, swimming coaches still need a measurement device that is accessible, easy to use, and provides easy-to-understand results. Conventional analysis systems such as high resolution cameras, while accurate, are too time consuming and cumbersome for daily use. With the advent of wearable sensors, especially inertial measurement units (IMU), motion analysis has gained the ability to not only study new aspects of motion, but also cover in-field applications. IMUs have been used to study swimming and provided a credible solution for extracting kinematic and spatio-temporal features. However, researchers focused primarily on extracting features rather than using them to evaluate performance and provide feedback in the field. Free-swimming is the phase that has been most studied with IMUs, and among the major swimming styles, front crawl has attracted more attention. However, to win competitions, swimmers need to master all phases of swimming in each style. Therefore, a comprehensive analysis approach based on IMUs covering all swimming styles and phases can provide the swimming community with deeper insight into swimmers' performance and a better solution to training needs. This thesis presents a novel methodology for swimming training analysis based on inertial wearable sensors. The proposed method uses a unified macro-micro analysis approach that scans the entire training session for swimming bouts and then narrows down to separating the swimming phases from wall to wall. The method is implemented using the IMU data from different sensor positions on the swimmer's body for comparison. Based on the results of the developed algorithms, sacrum position was determined to be optimal for detecting all swimming phases. As a result, the analysis then detects a set of spatio-temporal and kinematic parameters based on the IMU data on the sacrum at each phase, which are used for the swimmers' phase based performance evaluation. Furthermore, the extracted parameters were shown to reflect different aspects of the swimmer's performance, such as propulsion, posture, or efficiency. The developed performance evaluation method estimates a set of velocity-based goal metrics, validated against the reference system, that represent how well the swimmer performed during the corresponding phase. These goal metrics can provide coaches with knowledge-of-result feedback so that they can focus on the swimmers' weaknesses and guide them more efficiently. Finally, the system is used to provide weekly feedback to a team of young swimmers to evaluate the application of feedback in practice and its impact on swimmers' progress. Overall, this thesis proposes a new approach to swimming performance evaluation based on IMUs, with a broader scope of application to all phases and styles of swimming. It aims to expand the application of IMUs in swimming by providing spatio-temporal and kinematic parameters that represent critical aspects of swimming and objectively evaluate swimmers' performance. The estimated goal metrics are sensitive to swimmers' progress during weeks of training. In addition, the potential of such an analysis system for in-field training sessions is tested and showed promising results. Finally, coaches can obtain a more detailed view of swimmers' techniques at both macro and micro levels using a single IMU sensor on a daily basis.
© Copyright 2022 Veröffentlicht von École polytechnique fédérale de Lausanne. Alle Rechte vorbehalten.