Human Activity Recognition Using Smartphone

Abstract

Smart devices like smartphones, smartwatches have made this world smarter than any other time at every scale. A lot of facilities can be taken from these devices. Proper use of built-in sensors such as accelerometer, gyroscope, GPS is a few of them. In everyday life, people do a lot of physical activities which can be important for analysis like health state prediction, how much exercise they do etc. by using those sensors based on Artificial Intelligence. In this paper we have implemented both machine learning and deep learning to detect and recognize eight activities with a maximum of 99.3% accuracy. Of those activities few are similar in physical movements and actions like sitting in a chair at home, standing, and sitting in a car. These are almost similar and difficult to distinguish. Going upstairs and downstairs are also almost similar to separate. So we showed that with more sensors and data collection points a wide range of activities can be recognized and the accuracies can be increased. We proved our point by comparing the results of using fewer sensors and again using data of only one position of either pocket or wrist. Then finally we showed that by putting all the sensors and data of pocket, wrist together, we can recognize those activities accurately and in this way, a wide range of activities can be recognized with precision.