Gait-Enhance:Robust Gait Recognition of Complex Walking Patterns Based on WiFi CSI

WiFi-based gait recognition is promising for applications in smart home, human-computer interaction, intrusion detection and other fields, and has attracted wide attention from researchers. However, due to the extremely high sensitivity of channel state information (CSI) to changes in the environment, the gait information captured by CSI changes when walking on different tracks. Existing WiFi-based gait recognition systems require the subject to walk on a fixed single track, which is the key factor limiting the application of the system. In order to achieve complex walking patterns for gait recognition including different walking tracks and directions, a robust gait recognition system Gait-Enhance is proposed. Gait-Enhance can achieve high accuracy gait recognition without restricting the user to walk in a single track. Firstly, a CSI reconstruction algorithm is designed mitigate the negative impact on recognition accuracy caused by different patterns of CSI changes in different directions of walking. Secondly, a key feature enhancement algorithm is designed to enhance the important role of key features in the recognition process by using the correlation of CSI change between different walking tracks. In addition, a CNN-based recognition model is constructed to learn gait features and achieve gait recognition for complex walking patterns. Gait-Enhance is evaluated on a publicly available datasets containing multiple walking tracks and each track is walked in both directions. Extensive experiments show that Gait-Enhance can achieve a stable accuracy of 97.11% to 98.12% on a datasets of 4 to 9 people.