Al. L. Ronzhin - Ph.D. (Eng.), Research Scientist, Laboratory of Speech and Multimodal Interfaces, St. Petersburg Institute for Informatics and Automation of the Russian Academy of Sciences (SPIIRAS). E-mail: ronzhinal@iias.spb.su

At designing of intelligent rooms for meetings, lectures, scientific and educational events the following methods of audio- and video signals processing are used most often: 1) detection and tracking of users based on video monitoring; 2) determination of the position, recognition and identification of their faces; 3) sound source localization; 4) speech recognition; 5) identification of a speaker; 6) speech synthesis. The paper presents a method for automatic registration of the participants, based on multistage procedures for monitoring zones of the room, which provides detecting of participants position, recording audiovisual data, which are needed for identification of them. Experimental evaluation of the method was carried out on a database with more than 55 thousand photos of 36 participants. During experiments three methods of face recognition LBPH, PCA and LDA were compared. Conditions at the experiments carrying out (different distances from the camera to a participant, lighting, mobility of participants while taking pictures), affecting on quality and quantity of extracted facial features from an image, which is directly effect on the accuracy of recognition and occurrence of false positives, the LBPH method shows best result 79.5% was achieved by using the LBPH method and the lowest false positive rate 1.3% was obtained with the PCA method.