NeuralWiGait: an accurate WiFi-based gait recognition system using hybrid deep learning framework

WiFi-based human authentication systems are garnering substantial attention for its non-intrusiveness, privacy-preserving, and cost-effectiveness. Identity recognition in a WiFi sensing is typically achieved by analyzing the Channel State Information (CSI) that is generated as people walk. However, existing systems largely rely on models that extract an individual feature, leading to suboptimal accuracy. To address this issue, we propose a novel WiFi-based gait recognition system(NeuralWiGait), which authenticates identities by automatically learning the gait features of various users. A data preprocessing scheme is first applied, effectively reducing the signal noise and complexity of the CSI samples. In particular, a new hybrid deep learning framework (WiGaitNet) is used for automatic feature extraction for WiFi-based gait recognition. WiGaitNet integrates a specifically designed convolutional neural network (CNN) with a Bidirectional Gated Recurrent Unit(BiGRU), capable of extracting spatial and temporal features from human gait CSI samples. Subsequently, the concatenated features are fed into a softmax classifier for identification. Experimental results on public datasets (Widar 3.0 and NTU-Fi-HumanID) show that the proposed system achieves an average accuracy of 99%, demonstrating tremendous potential for application.