Multimodal Wearable Sensing for Sport-Related Activity Recognition Using Deep Learning Networks

Abstract

Wearable sensors using sensor-based Human Activity Recognition (S-HAR) are generally capable of regular simple actions (walking, sitting, or standing), but are indistinguishable from sophisticated activities, such as sports-related activities. Because these involve a more comprehensive, contextual, and fine-grained classification of complex human activities, simplex activity recognition systems are ineffective for growing real-world applications, for example remote rehabilitation observation and sport performance tracking. So, an S-HAR framework for recognizing sport-related activity utilizing multimodal wearable sensors in numerous body positions is proposed in this study. A public dataset named UCI-DSADS was used to investigate the recognition performance of five deep learning networks. According to the experimental results, the BiGRU recognition model surpasses other deep learning networks with a maximum accuracy of 99.62%.