Detection of swimming stroke start timing by deep learning from an inertial sensor

We aim for developing a system for supporting swimmer with an inertial sensor. It is required to develop a method to detect swimming stroke starting timing from the single inertial sensor data attached to swimmer’s back waist (3-axis acceleration and gyro data). In this paper, we propose a detection method of swimming stroke starting timing from the time and frequency domain of 3-axis acceleration and gyro data by using the deep learning and peak detection. To learn the deep learning parameter, we carried out an experiment. The number of subjects is six and they swim butterfly. We assigned four persons as the training dataset and two person as the test dataset. As a result, our method achieves the high quality detection, i.e., the precision and recall are 0.855 and 0.904, respectively. Therefore, we confirm that there is a possibility that our method can use our system for supporting swimmer with an inertial sensor. Moreover, our method can be used to an estimation of starting timing of other human motions.