Research Of Stereo Matching Based On Point Feature

Face recognition is a technique of personal authentication, which can be extensively applied in fields of security control. Because 2D face recognition is easy to be influenced by intensity, expression and so on, and 2D facial images provide so little information that precision of recognition cannot be enhanced, 3D face recognition is the tendency of face recognition, and 3D-face-model is needed. Since acquiring 3D information directly with equipment, such as 3D scanner etc., costs much and needs to store mass datum, head modeling can be achieved effectively by stereo vision which is a technique of obtaining deepness based on two or many 2D images. Stereo matching based on point feature is the key of head modeling based on 2D image. Stereo matching based on point feature is the kernel in stereo matching. After analyzing existing technique of stereo matching based on point feature, Results of research in this paper can be summarized as follows:①Parallel binocular vision system is introduced and its properties are proved. Composition and principle of parallel binocular vision system are introduced. Two characteristics that Epipolar lines coinciding with the horizontal scanning lines and corresponding points in both images lying on the corresponding epipolar lines are proved. This kind of vision system is the base of acquiring 3D position of the spacial points based on stereo matching.②Direct method of stereo matching based on point feature is acquired. Edge point feature is mainly utilized, existing methods of extracting and matching edge points are analyzed and correlative experiments are performed. Traditional Barnard's algorithm is improved by discrete dyadic wavelet transform, and results of our experiment prove that inaccurate matching and un-matching feature points are reduced and speed of matching is raised. In addition, during the research, the author finds that discrete dyadic wavelet transform is used for transforming digital image which is acquired by sampling the sequential image with the same sampling frequency in two dimensions and it is NxN . Actually, most images are N×M (N ≠ M),so, discrete dyadic wavelet transform used for transforming digital image f(x,y) is studied and feasibility of the proposed algorithm is proved in theory by related deduction.③Indirect method of point stereo matching based on line feature is explored. Indirect method of stereo matching based on point feature is introduced in detail.Acquiring and matching of edge lines and how to match edge points by edge lines matching are mainly discussed. Related experiments are performed and results of the experiments prove that it is effective to match points utilizing line feature.