Person re-id through radio biometric signatures, human silhouette and skeleton video synthesis through wi-fi signals

The increasing availability of wireless access points (APs) is leading to human sensing applications based on Wi-Fi signals as support or alternative tools to the widespread visual sensors, where the radio signals enable to address well-known vision-related problems, such as illumination changes or occlusions, and human privacy concerns. These are possible because both objects and people affect wireless signals, causing radio variations observed through the channel state information (CSI) measurement of Wi-Fi APs that allows signal-based feature extraction, e.g., amplitude or phase. On this account, this thesis shows how the pervasive Wi-Fi technology can be directly exploited for solving person Re-ID, for the first time in literature, and image synthesis problems. More accurately, for the former, Wi-Fi signals amplitude and phase are extracted from CSI measurements and analyzed through a two-branch deep neural network working in a siamese-like fashion. The designed pipeline can extract meaningful features from signals, i.e., radio biometric signatures, that ultimately allow the person Re-ID. The effectiveness of the proposed system is evaluated on a specifically collected dataset, where remarkable performances are obtained; suggesting that Wi-Fi signal variations differ between different people and can consequently be used for their re-identification. Instead, for the image synthesis, a novel two-branch generative neural network effectively maps radio data into visual features, following a teacher-student design that exploits cross-modality supervision. This strategy conditions signal-based features in the visual domain to completely replace visual data. Once trained, the proposed method synthesizes human silhouette and skeleton videos using exclusively Wi-Fi signal amplitude. The approach is evaluated on publicly available data, where it obtains remarkable results for both silhouette and skeleton videos generation, demonstrating the effectiveness of the proposed cross-modality supervision strategy. Concluding, a small use case is specified to show how videos synthesized from wireless signals can be used to solve human activity recognition obtaining a privacy-conscious system and potentially exploiting the radio signal benefits.