Through-Wall Human Motion Recognition Using Random Code Radar Sensor With Multi-Domain Feature Fusion

Abstract

A random code radar sensor with multi-domain feature fusion is proposed and demonstrated experimentally for through-wall human motion recognition. This radar sensor transmits the aperiodic physical random code signal generated from the Boolean chaos as the probe signal, which can realize the anti-interference and unambiguous detection. The slow time-range (ST-R), slow time-Doppler frequency (ST-DF), and range-Doppler frequency (R-DF) maps are obtained by the correlation ranging, short-time Fourier transform (STFT), and fast Fourier transform (FFT), respectively. Furthermore, the features of ST-R, ST-DF, and R-DF domains are extracted by the fast principal component analysis, and fused by a simple serial connection to realize the complementarity and perfection of multi-domain features. Finally, the motion recognition is implemented with the support vector machine multi-classifier. The results show that utilizing the anti-interference random code signal and serial feature fusion, the proposed radar sensor can accurately recognize the human motions behind the wall without massive calculation and high computer configuration, and the average accuracies reach 97.22% for ten motions and 93.08% for six volunteers. Compared with the features of single domain and two domains, the multi-domain feature based on serial fusion can improve the recognition accuracy quickly and efficiently.