A 2-Stage Para-Hermitian Eigenvalue Decomposition for Crowd Counting in an Indoor Setting Based on the Scattered Radio Field

Frank E. Ebong, Andrea M. Tonello

Abstract: In this paper, we propose a Sliding Window 2-Stage Para-Hermitian Eigenvalue Decomposition (PhEVD) algorithm for extracting the scattering amplitudes of a scattered radio field from a high-dimensional and multichannel, time-varying Wi-Fi Channel State Information (CSI) data in a multi-antenna setup. The algorithm is designed to address the computational challenges associated with processing large, high-dimensional and multichannel Wi-Fi CSI data. We evaluate the proposed method on real-world Wi-Fi Channel State Information (CSI) data collected in an indoor environment, considering seven distinct scenarios ranging from an empty room to a room with up to six people. The extracted scattering amplitudes are used to classify these scenarios using a fine-tuned Gaussian Support Vector Machine (SVM). Our experimental results demonstrate that the proposed 2-Stage PhEVD algorithm outperforms state-of-the-art feature extraction methods, such as Reconstruction Independent Component Analysis (RICA) and sparse filtering, in terms of classification accuracy of the scattered radio field. This work highlights the potential of the proposed approach for device-free sensing, non-intrusive crowd counting and other applications in indoor sensing.

External IDs:dblp:conf/eusipco/EbongT25