Edge Computing Enabled Human Activity Recognition (ECEHAR) Using LSTM and CNN

Abstract

Human Activity Recognition (HAR) is an important research area for various application domains such as healthcare, gaming, telemonitoring, and sports. However, executing HAR algorithms on remote servers or in the cloud have challenges in terms of high latency, increased bandwidth demand, and high energy consumption. Moving the computation to edge-assisted HAR is more effective and flexible solution to address the limitations of conventional approaches. In this paper, a set of salient points are identified on the human body and are represented mathematically as triangles. Human activities affect the angles of the triangle, and the resulting deformation is used for classifying the activity. Both Convolutional Neural Network (CNN) and Long Short-Term Memory (LSTM) are used for human action classification and have good performance with accuracy as 99.8%. The performance of Edge Computing Enabled Human Activity Recognition (ECEHAR) is evaluated on both benchmark and real-time datasets using precision, recall, F1-score, and accuracy. The model has shown promising results compared to contemporary methods.