2025 年 16 巻 3 号 p. 506-518
Wi-Fi sensing has emerged as a cost-effective and privacy-preserving solution for motion detection, offering significant potential for applications in intrusion detection and smart environments. This study proposes a novel preprocessing approach to enhance motion detection accuracy by addressing two critical challenges in Channel State Information (CSI) data: amplitude biases in the subcarrier direction and high-frequency noise in the temporal domain. The preprocessing methods—removal of DC components and lowpass filtering—are designed to isolate meaningful patterns while suppressing noise that could mislead classification. A deep learning architecture integrating Convolutional Neural Networks (CNN) and Long Short-Term Memory (LSTM) networks is employed to extract spatial and temporal features from preprocessed CSI data. Experimental evaluations demonstrate that the proposed system performs better than existing methods, particularly in detecting small movements, with a 17.4% improvement in true negative rates and enhanced robustness to environmental variations.
