Pattern-Independent Human Gait Identification with Commodity WiFi

Recent years have witnessed the significant advancement of WiFi-based human gait identification. However, existing works require the human subjects to walk with a standard gait pattern, i.e., the normal walking speed and free limb swings. This significantly impedes the wide adoption of these technologies since humans walk with diverse gait patterns in reality. To this end, we present a Pattern-Independent Authentication System (PIAS), the first system that enables human identification across different gait patterns using commodity WiFi. First, we extract Doppler spectrograms as human gait signatures and standardize them to narrow the distribution discrepancies between different gait patterns. Then, we design a Class-level Unsupervised Domain Adaptive Network (CUDAN) to extract human identity features in the source and target domains while minimizing the feature domain discrepancies at the class-level to achieve pattern-independent human gait identification. Extensive experiments are conducted on different gait patterns and the results demonstrate that our system can achieve the highest performance compared to the baseline methods, demonstrating its effectiveness in human identification across diverse gait patterns.