WiFi CSI Based Passive Human Activity Recognition Method Using BLSTM-CNN

Abstract

Human activity recognition (HAR) based on WiFi channel state information (CSI) has been received strong attention in recent years. For this problem, vision-based and sensor-based approaches can provide better data at the cost of user inconvenience and privacy issues. In contrast, radio frequency-based approaches for passive sensing without devices typically employ received signal strength indicators (RSSI), ignoring the potential benefits of fine-grained sensing accuracy of CSI. Deep learning is considered one of effective approaches for designing HAR method in WiFi CSI environments. Hence this paper proposes a deep learning scheme based on the CSI of WiFi, which uses a combined network of bi-directional long short term memory (BLSTM) and convolutional neural networks (CNN) for passive HAR using WiFi CSI signals. BLSTM can effectively use the input forward and backward feature information, and then the CNN layer to achieve passive human activity recognition. Simulation results show that the method significantly improves recognition accuracy compared to most current recognition methods.