Wi-Fi CSI-Based Human Presence and Respiration Detection in Non-Line-of-Sight Environments Using Dual-Receiver Sensor Fusion on Edge Devices

Abstract

A fundamental challenge in search and rescue (SAR) operations is detecting victims trapped beneath collapsed structures under non-line-of-sight (NLOS) conditions, where building materials obstruct the direct propagation path between the sensing system and the victim. This paper presents a proof-of-concept Wi-Fi Channel State Information (CSI) system for detecting human presence and estimating respiration rate under controlled NLOS conditions using ESP32-S3 N16R8 devices. The system employs one transmitter and two receivers positioned outside a rigid pressed-wood partition to enable confidence-weighted fusion of two CSI links. CSI amplitudes are processed through resampling, z-score normalization, detrending, band-pass filtering, top-subcarrier selection, weighted fusion, and Fast Fourier Transform (FFT) peak analysis. Human presence is inferred from dominant-peak spectral confidence, while respiration rate is estimated from the corresponding peak frequency. In six controlled 120-s trials, the system correctly separated three empty-room trials from three human-present trials, yielding an accuracy, precision, recall, and F1-score of 1.000. Across the three human-present trials, respiration-rate estimation achieved a mean absolute error (MAE) of 0.319 BPM, a root mean square error (RMSE) of 0.515 BPM, and a bias of -0.319 BPM. These results support the feasibility of ESP32-S3 N16R8-based Wi-Fi CSI sensing for controlled NLOS conditions relevant to SAR-oriented sensing.