Swimming motion classification for coaching system by using a sensor device

We are engaged in developing a swimming motion coaching system by using a sensor device. One of the requirements of the system is the process of automatically estimating and dividing the section of swimmer’s motions (such as stroke and turn) from the sensor data. In this paper, we proposed a method of estimating the section of each swimming motion in four swimming styles (front crawl, backstroke, breaststroke and butterfly). A classifier of swimming motions based on decision tree was constructed by learning data. As a verification in the generalization ability of the classifier by test data, F-measure was ≥. 713 for all swimming styles. We also estimated the start and end points of the section of swimming motions. The estimated mean errors of the start and end points of turn in all swimming styles were ≤. 488 seconds (except for backstroke) and ≤. 514 seconds, respectively. From the pattern recognition point of view, we found that we could classify the features of stroke and turn in four swimming styles. However, from the user’s point of view, we should aim to achieve much higher accuracies.