Cell phone based exercise tracking system

Abstract

In the past, many research studies of human activity recognition (HAR) had used body-worn sensors to identify users' physical activities. But, limiting users' mobility is the main drawback of this kind of approach. Nowadays, smartphones are equipped with different types of built-in sensors which the data generated by some of the built-in motion sensors are useful for identifying the physical activities of smartphone users without affecting their mobility. This project aims to implement a cell phone-based exercise tracking application for recognizing five types of user's physical activities. The raw data generated by the built-in accelerometer and gyroscope of a smartphone had been used for training a Long Short Term Memory (LSTM) model. As a result, the final model had around 81 percent of testing accuracy on identifying the user's workout type. After requiring the user to input the workout type before tracking for several times, the Flutter mobile application will be able to record the workout details and identify the workout types based on the results returned by the LSTM model.