Research of Visual Features Detection and Tracking Methods about Audio-Visual Bimodal Speech Recognition

Audio-visual bimodal speech recognition can improve speech recognition rate, the lip detection, location and tracking is the key of bimodal speech recognition system. This article discusses the lip detection, location and tracking algorithms of bimodal speech recognition. Locate lips precisely by use geometric structure of face, relative position of lips and separable color information of color space. Using adaptive color filter to segment the lip contour effectively, and use PMM algorithm to locate and track lip precisely. Experimental results shown that the algorithms studied in this paper can detect, locate and track lips precisely, robustly and quickly.