HDL: Hierarchical Deep Learning Model based Human Activity Recognition using Smartphone Sensors

With the development and popularization of smart-phones, human activity recognition methods based on contact perception are proposed. The smartphones which are embedded with various sensors can be used as a platform of mobile sensing for human activity recognition. In this paper, we propose an automated human activity recognition network HDL with smartphone motion sensor units. The HDL network combines DBLSTM (Deep Bidirectional Long Short-Term Memory) model and CNN (Convolutional neural network) model. The DBLSTM model is first used to model long sequence data and ultimately generate a bidirectional output vector in a abstract way. The DBLSTM model is good at dealing with serialization tasks but poor in the ability to extract features. Hence, the CNN model is then used to extract features from the abstract vector. Finally, the output layer employs a softmax function to classify human activities. We conduct experiments on the Public domain UCI dataset. The experimental results show that the proposed HDL network achieves reliable results with accuracy and F1 score as high as 97.95% and 97.27%. Compared with other networks based on the same smartphone dataset, the accuracy of HDL is higher than S-LSTM and Dropout CNN network by 2.14% and 6.97% respectively.