The Implementation Of Lip Movement Acquisition System Based On Adaboost And Optical Flow

Extraction of movement of lips in the Video is a valuable topic, and it has practical application value in the field of human-computer interaction, facial recognition and synchronization between audio and video. Its research has both theoretical and practical significance.In this paper, techniques and methods to extract movement of the lips were studied. Methods based on optical flow and Adaboost cascade classifier were proposed for the extraction. First, we used images with lips to train the Adaboost cascade classifier. Secondly, establish lip color model in the YCgCr color space for detection and location of the lip area, and then we got points on both sides of the mouth. Finally, track subpixel corner points obtained by the points using optical flow method, and get the curve of lip movement. The main contributions of this paper are the following two aspects:(1) We realize the combination of Adaboost cascade classifier and YCgCr color space model to improve the accuracy of Adaboost classifier.(2) We propose a method of tracking sub-pixel points by using optical flow, which could get the curve of lip movement.In addition, the initial points are getting from the perpendicular bisector of the line connecting both sides of the mouth.