Cross-Time Domain Dynamic Human Detection Based on WIFI-CSI Feature Weighted Voting

To date, numerous studies have been conducted on passive sensing techniques for WiFi-CSI, yet the application of these methods to real-world scenarios still faces challenges across both temporal and spatial domains. For dynamic human body detection across the time domain, this paper proposes a featureweighted voting (FWV) algorithm, which integrates the prediction results of various features to achieve a comprehensive prediction. First, the collected CSI data is processed by calculating the amplitude and applying denoising techniques. Based on this, five CSI features are extracted, each characterizing different aspects of the dynamic human body’s CSI data. Then, the trained model is used to predict the validation set and the weights of each feature are calculated based on the prediction results of the validation set. Finally, by combining the feature weights with the respective prediction results, a comprehensive decision is made to determine the presence of a dynamic human body. Experimental results demonstrate that the proposed method achieves an accuracy of over $99 \%$ in dynamic human body detection, validating the effectiveness of the approach.