WiZeCSi: Toward Wi-Fi-Based Single-Link Zero-Effort Cross-Domain Gesture Recognition

Currently, much effort is devoted to improving the accuracy of Wi-Fi sensing. Despite the typically high density of modern Wi-Fi deployments, in many scenarios, the number of access points under control is limited, which does not allow the simultaneous usage of multiple Wi-Fi links for sensing. To address this issue, the paper proposes a Wi-Fi-based Single-link Zero-effort Cross-domain gesture recognition approach called WiZeCSi. A key feature of WiZeCSi is multi-ratio augmentation and link mixing methods that maximize the single-link sensing ability, converting Channel State Information to Single-Link-Single-Ratio samples and expanding the dataset. In the evaluation with a popular Widar3 dataset, the WiZeCSi achieves state-of-the-art results with 98.3% in-domain accuracy and higher than 96.6% cross-domain accuracy in single-link sensing. The designed WiZeCSi approach also enables sensing with an arbitrary link combination and cross-link sensing to overcome the link switching and device relocation problems.