Physique- Based Human Activity Recognition Using Deep Learning Approaches and Smartphone Sensors

Abstract

Understanding human actions via the analysis of sensor data captured by wearable sensors is the goal of the complex subject of study known as sensor-based human activity recognition (S-HAR). Human participants' characteristics are only periodically included in deep learning (DL) approaches to S-HAR. Recognizing people was challenging for these DL methods because of the variety of physical characteristics people have. To address this challenge, we introduce a physique-based S-HAR architecture that could support deep learning networks to achieve higher identification a ccuracies a nd F1-scores. The HARSense dataset, a publicly available benchmark S-HAR dataset that compiles raw sensor data acquired from smartphones, was employed to build and evaluate five DL networks. A ccording to the experiments, the five models' detection performance improves dramatically when given access to biological data.