Human Activity Recognition using Deep Learning

Abstract

Human activity recognition (HAR) is a difficult task that entails recognizing and categorizing human actions using sensor data gathered by mobile phones and wearable. Because of recent developments in deep learning, deep neural networks (DNNs) have surpassed other methods as the gold standard for HAR. In this study, we provide a high-level introduction to HAR by making use of deep learning methods. We begin with an overview of the HAR problem and its obstacles, such as the inherent uncertainty of human behaviour and the enormous dimensionality of sensor data. Then, we talk about how deep learning plays a part in HAR and provide a thorough assessment of recent studies that have used several kinds of deep learning architectures, such as CNNs, RNNs, and hybrid architectures, to solve HAR problems. We also discuss some of the important issues and on-going research questions in HAR utilizing deep learning, including how to handle imbalanced datasets, how to interpret DNN models, and how to cope with the sparsely and variability of sensor data. We wrap up by talking about where this type of deep learning-based HAR could go in the future, including applications like personalised care monitoring, activity- based intelligent assistant, and smart home automation.