An Automatic Feature Selection Model for Running Sprint Reorganization Using Wrist Accelerometers

For the existing motion recognition tasks, few researchers have studied running sprint, that is, the sprint part of the running segment is discriminated. In order to monitor the running situation of sportsman, we proposed an Automatic Feature Selection Mode (AFSM) for running sprint reorganization using wrist accelerometers. First, information such as acceleration signal is collected through wearable devices. After pre-processing and feature extraction, an automatic feature selection algorithm is used to select important features. We also propose two key features, heart rate and acceleration derivatives, for experiment. The experimental results show that AFSM can automatically screen useful features to achieve better classification accuracy, effectively identify the sprint period in the running segment, and the recognition accuracy rate reaches 83.7%; Heart rate and acceleration derivatives are the key information for sprint recognition, which can significantly improve the classification effect, improve the accuracy rate by 1.77
^%
, and achieve more accurate recognition.