End‐to‐end motion segmentation and recognition with high accuracy using wearable devices

Abstract

Abstract An important research direction in applying wearable devices in sports is motion recognition and segmentation. The authors propose an end‐to‐end motion recognition–segmentation method in basketball. First, data are collected using an arm guard with an inertial measurement unit (IMU) module. Next, a sliding window is used to extract the data features. The feature extraction method is a multi‐model long short‐term memory (LSTM) network. After feature extraction, a label vector is obtained and a result vector is formed after temporal matrix processing. The results of the experiment on the accuracy, precision and recall were 97.1% ± 1.6%, 94.4% ± 0.7% and 96.4% ± 1.1%, respectively, showing better performance among similar studies.