The Study Of Terrestrial Stereo Matching Techniques

Stereo matching is a fundamental aspect of many problems in digital photogrammetry and computer vision. It is used in many fields and the development of it has decided real-time and degree of automation of digital photogrammetry.The content of research is mainly about terrestrial stereo matching technique in this paper. The theory of matching algorithms and processes are studied and analyzed deeply. A testing system has be established to check the approached methods. By reviewing and summarizing the existing stereo matching technique, the structure of matching system has been accessed and a novel matching system based on feature points has also been proposed. Then every integral part of it has been expounded. First, some familiar operators of feature points extraction are analyzed. The feature points with good distribution are extracted by using of a improved Harris operator. They are good for the matching on the next. Second, the matching system is described on matching measure, constraint condition and matching strategy in this paper. The matching measure is Euclid-distance in the two grads vectors. They are related to feature points and established by using a isotropy modular. The constraint condition includes epipolar geometry, affine transform and uniqueness. The three constraints are used for checking the matches with different combinations and they complement each other well. The two-level matching strategy is used for avoiding global searching on large image and improving accuracy and efficiency.The experiment is about different kinds of terrestrial images and provides that the matching system of this paper is not sensitive to noise and guarantees better reliability and accuracy. For images with rotating and view point changing, it can also achieve good matching effect. Moreover, it is compared with NCC and SIFT. The result shows that the matching accuracy of using the method of this paper is better than using other two and the matching speed is higher than NCC. Besides, the matching result is used in auto-building DSM, image mosaic and stereo display. The result shows that these applications are success.Finally, the key work of this paper is summarized and the outlook is given.