Framework for Human Activity Recognition

Abstract

Human Activity Recognition (HAR) is a critical component in many applications, including health tracking and human survey systems. It is also a hot research subject in health care and smart homes. Also, Smartphones became an integral part in everyone’s life. The majority of people use it to search for information, watch movies, play games, and connect with their social networks, but there have been many useful studies on smartphones. Smart phones come with a range of sensors built in, including an accelerometer, gyroscope, GPS, compass sensor, and barometer. Using these sensors, a HAR device can be built to capture the user's condition. We can reliably detect the user's behaviour by implementing machine learning techniques into the system. The proposed work uses the LSTM algorithm to estimate human motion movement using raw sensor readings from mobile sensors in our smartphone as inputs.  With minimal training data and usable memory in smart phones, the proposed work recognises the user's behaviour. The user's behaviours, such as walking, running, lying down, sitting, and standing, can all be observed