HHARNet: Taking inspiration from Inception and Dense Networks for Human Activity Recognition using Inertial Sensors

Abstract

Human Activity Recognition (HAR) is an important area of research in the light of enormous applications that it provides, such as health monitoring, sports, entertainment, efficient human-computer interface, child care, education, and many more. The use of Computer Vision for Human Activity Recognition has many limitations. The use of inertial sensors which include an accelerometer and gyroscopic sensors for HAR is becoming the norm these days considering their benefits over traditional Computer Vision techniques. In this paper, we have proposed a l-dimensional Convolutions Neural Network which is inspired by two state-of-the-art architectures proposed for image classifications; namely Inception Net and Dense Net. We have evaluated its performance on two different publicly available datasets for HAR. Precision, Recall, Fl-measure, and accuracies are reported.