Bidirectional Gated Recurrent Units For Human Activity Recognition Using Accelerometer Data

Human activity recognition aims to detect the type of human movement based on sensor data gathered during human activity. Time series classification using deep learning approaches offers opportunities to avoid intensive handcrafted feature extraction techniques where the efficiency and the accuracy are heavily dependent on the quality of variables defined by domain experts. In this paper, we apply recurrent neural networks on data collected from mobile phone accelerometers for the recognition of human activity. More specifically, we use the bidirectional gated recurrent units mechanism. The results show that this technique is promising and provides high quality recognition results.