The purpose of this study was to quantify and qualify the swimming performance using a wearable wrist mounted inertia sensor and data mining analysis. The authors measured swimming stroke motion with tri-axial accelerometer and gyroscope during 50m interval training workout of sixteen collegiate competitive swimmers with four swimming styles. The authors examined four quantitative parameters such as swimming time, stroke rate, stroke counts, swimming distance (number of bouts). The obtained sensor signal data showed two states, such as resting and non-resting (swimming) during the work-out. Non-resting state includes leaning forward, water start, turning, gliding and stroking phases in it. Firstly, the authors identified those two states by using the acceleration. Then, stroking phase was identified by using an optimum threshold. Resting/non-resting state identification resulted in the estimation of the swimming time. As for the freestyle (FR) and butterfly (BU), the successful result ratios of the stroke time estimation compared with manually measured swimming time were r2 = 0.896 and r2=0.874 respectively. On the other hand, those of backstroke (BA) and breast stroke (BR) obtained lower ratios, because of the failure of the start and goal events identification. A multi-layered neural network (MNN) and C4.5 decision tree were adopted for the classification of the stroke styles. The successful classification of the four stroke styles by MNN was 100% (with acceleration and angular velocity) and 98.6% (with acceleration only). C4.5 decision tree could classify four strokes with 96% (with acceleration and angular velocity) and 94.7% (acceleration only). The stroke count was estimated by the threshold method then converted into binary stroke signal pattern (0 or 1). A cluster of continuous binary stroke signal means one-way swimming of the each bout. Therefore, we could estimate accumulate swimming distance as well. Successful estimated distance on FR was 78.9% and 76.3% for BU. Those estimated parameters can explain both of intensity and total volume of the swimmer`s training. Although the further investigation must be necessary, those can be used for the basis of estimation of the energy expenditure during the training session in the future.
© Copyright 2018 XIII th International Symposium on Biomechanics and Medicine in Swimming Proceedings. Published by Impress R&D. All rights reserved.