Human Activity Recognition based on Smartphone Sensors- A Comparative Study

Abstract

HAR has become a leading area of research because of its noteworthy contribution in applications that aim to improve the quality and standard of life. HAR system also contributes to health and safety in smart cities, privacy and security, etc., which directly or indirectly improves the quality of service towards society. In this study, we studied the different techniques used for the detection of human activities using built-in sensors in smartphones. In all these techniques raw data is collected using gyroscope and accelerometer sensors inbuilt in the smartphones and then different data preprocessing steps are implemented to clean the data. Important features are extracted using different feature extraction techniques. Finally, the “Machine Learning” or “Deep Learning” models are trained which can accurately recognise the activities. We analyze several modern deep learning techniques which provide excellent results due to their capability of learning deep features. Also, we have analyzed the research gaps in the current literature which provides a sound understanding to identify the future work required in this area of research.