ADAPTIVE NON-LINE-OF-SIGHT IDENTIFICATION SYSTEM BASED ON BEST FEATURE SUBSET

Identification has always been the focus of research in the field of security, and its research in non-line-of-sight scenarios is of great significance. Aimed at comfort and privacy of recognition, a best feature subset based adaptive non-line-of-sight identification system is proposed. Low-dimensional useful data of Wi-Fi signals was obtained by effectively combining multiple preprocessing methods. A robust human detection method was proposed to intercept effective fragments. A supervised feature extraction method was designed, and "forward search" was employed to obtain the best feature subset. A traditional Adaboost algorithm was improved to realize adaptive recognition under group variation. Experimental evaluation shows that when the number of volunteers in system is 2~12, which has better performance compared with related systems and traditional classification algorithms.