Feature Points Extraction And Fast Matching Of Stereo Vision Images

In the human existent environment, vision is a significant way to acquire information and recognize the world. Computer vision makes the computer cognize the three-dimensional environment by the two-dimensional images, including its shape, position, posture, movement and so on. Moreover can carry through the description, memeory, recognition and understanding to them. Computer vision technology is widely used in each aspect, it can be said that occasions which need the human vision all need the computer vision.Feature points extraction and matching of stereo vision images is a key technology and research focus in the computer vision field. Research of feature points extration and matching have an important role and significance for face recognition, three-dimensional reconstruction and image splicing.This papar have finished a deep theoretic study and sufficient practical works. It summarizes the basic knowledge, including the concept of digital images and feature points, stereo matching principle, parallax ranging principle, matching feature choice, matching constraints and geometric theory. It summarizes five kinds of feature points extration algorithms base on gray, compares and analyses those five methods by experiments. SIFT feature points extraction method is described detailedly in this paper, also it validates the accuracy and effectiveness of the algorithm by experiments. For the stereo matching, the traditional matching algorithms are classified in this paper, it emphasizes the quasi-dense matching method and improves it in two aspects. First, convolution is introduced in the calculation of normalized correlation coefficient to enhance the speed. Second, realizes adaptive search window by using the confidence coefficient to enhance the precision. The experimental results show that the quasi-dense matching method which improved in this paper greatly enhances the validity and veracity.