User Location Estimation Based on Stereo Audio-Vision

Two methods have been used to track user location: stereo audio-based and the stereo vision-based methods. The stereo audio-based method is sensitive to noise, its performance deteriorates readily. The stereo vision-based method has problems in estimating location when the surrounding environment is restricted or the user is out of camera sight. In this paper, we compensate for the problems of the two methods. We used GCC-PHAT for audio-based location and the skin color and the AdaBoost algorithm for vision-based location adopting disparity information and the stereo camera internal transformation of the detected image. From the experimental results, the error of the estimated average angle is 1.32 and 0.86 degrees at 1 and 2 meters, and we also confirm that the proposed method is very efficient in user location estimation of near distance.