INTRODUCTION: The aim of the present study was to propose an automatic temporal phase detection, such as water hand entry (t_en), beginning of pull (t_pull), beginning of push (t_push) and start of aerial recovery (t_rec), and inter-arm coordination estimation in front-crawl swimming using IMUs for paraswimmers with physical impairments. We examined the validity of our method by comparison against a video-based system, considered as the reference technique to estimate those key points throughout the trials. To prove the efficiency of the proposed method, the proposed algorithm was performed on a Paralympic swimmer to compute the key points of the stroke in front crawl used as a case study to compute the Index of Coordination (IdC). METHODS: Participants represented 5 of the International Paralympic Committee impairments categories (S8, S9, S10, S12 and PTS4). We performed 4 trials over 25 m: one was performed at slow speed / one was performed at fast speed / one was performed in breath holding condition at maximum speed / one was performed with preferred breathing condition at maximum speed. The proposed data treatment procedure, using Matlab, aims to compute different stroke phases of front-crawl and paraswimmers coordination by measuring various key events as described in Chollet et al. (2000). The results of key points detection are obtained from the angular velocity signal. RESULTS: Linear regressions which were close to the identity line (slopes of the regressions all between 0.99 and 1). Bland and Altman statistics (mean bias ± LoA) revealed on average that the mean bias and the limit of agreements values were equal to -0.01 ± 0,05 for t_en, 0.04 ± 0,152 for t_pull, -0.03 ± 0,105 for t_push and 0.04 ± 0,114 for t_rec. >From these results we can see that the mean bias was not increased by the physical impairments of the paraswimmers (the difference never exceeded ± 2 video frames of 0.02 s). The LoA were systematically higher for paraswimmers, especially for t_pull with values reaching ± 0.15 s. The RMSE values were 0.029 s for t_en, 0.087 s for t_pull, 0.06 s for t_push and 0.068 s for t_rec. RMSE results confirmed that the detection performance of the present algorithm is challenged for t_pull caused by the variety of our sample of paraswimmers. CONCLUSION: In the present work, we proposed an automatic algorithm that uses mainly the raw data extracted from IMUs positioned on the lower arms of the paraswimmers. This study has shown that the key points allow to estimate the coordination, and therefore the motor control, the motricity of swimmers despite the difference in swimming caused by the existence of physical impairment. We may therefore conclude that neither the swimming technique, nor the swimming kinematics negatively influence the general score of the algorithm that never roots its detections on constant values with thresholds but rather on slope changes or peak detections.
© Copyright 2023 28th Annual Congress of the European College of Sport Science, 4-7 July 2023, Paris, France. Veröffentlicht von European College of Sport Science. Alle Rechte vorbehalten.