Deep Learning based Human Activity Recognition using Smartphone Sensor Data

Abstract

Human activity recognition is one of the most trending domains in the field of computer science since past many years. The main task of Human activity recognition is to classify the actions performed by Human beings by analysing the data. These data can be captured either through a series of videos or data captured from sensors such as accelerometer, gyro meter, magnetometer, etc. There are wide range of application domains where this task can serve and help people to analyse the Human behaviour. Some of the application areas include healthcare and fitness, surveillance and analysis of video, education domain and many more. There has been a lot of work done on this problem in the past. This report will introduce the Human activity recognition, application domains, various ways of identifying the actions and challenges faced in current era. It will also discuss about some of the work done in the past and the drawbacks of them. The state-of-the-art dataset WISDM will also be discussed and analysed. Later on, we will introduce the approach we used to solve this problem and analysis of the proposed architecture. Finally, it will include results and analysis tested on the WISDM dataset using the proposed model.