Human activity recognition uses heterogeneous sensors to capture the state of user and environment. This has wide use in heath-related application such as health monitoring and athlete training. This project proposes a promising human activity recognition approach based on long-short term memory (LSTM) method using the smartphone inertial sensor data. Different kinds of network configuration are tested and explored in our experiment. A comparison with traditional machine learning method is also given in this paper.
Structure of the recurrent neural network model.