Research on Gesture Recognition Based on Wi-Fi Signal

CSI-based gesture recognition has received wide attention in the field of intelligent perception due to its non-contact and penetrating nature. Gesture recognition using Doppler shift for CSI is affected by different sensing configurations resulting in large differences in spectral quality, and existing methods are still challenging in terms of cross-domain robustness. In this paper, a DFS feature enhancement method is proposed. First, two transceiver links are used to extract different DFSs for the same gesture, and the main gesture features are obtained by setting dynamic thresholds and eliminating the lower power and dispersed power of the DFS through power differencing; on this basis, the segmentation mechanism of the main spectrum and the sub-spectrum is constructed based on the changes in waveforms of the DFS and the power distributions of the main and sub-spectrum are analyzed to establish a power transfer model from the sub-spectrum to the main spectrum; after that, the main spectrum is subjected to power restoration to ensure the complete gesture features, and the enhanced primary spectrum is called DFS+; finally, the DFS+ is recognized using a deep neural network fused with CNN and RNN. The results show that DFS+ enhances the cross-domain gesture features while ensuring the completeness of the gesture features, and achieves an accuracy of 96.3 % in traditional gesture recognition and 92.1 % under the influence of multiple sensing configurations, which significantly improves the stability and cross-domain robustness of gesture recognition.