Three-dimensional Reconstruction Based On Stereo Pairs Of Metal Fracture

The recognition and classification of metallic fracture surfaces have certain guidance to improve equipments and consequently avoid the happening of rupture. Entire 3-D information of fracture surface, which provides fine basis for going deep into the understanding of microcosmic fracture process, can be obtained by making use of the technology of stereo vision. An algorithm of SEM 3-D surface measurement and reconstruction has been established on personal computer based on the research accomplishment of image processing and computer vision.The main research content of computer vision is reconstructing the3-D information from many pieces of 2-D images, and the binocular computer vision technology based on two pieces of images is a hotspot of research. The basal principle of binocular computer vision is copying person eye and human perception process, observing the same scenery from two viewpoints to gain images from different visual angle, and then gaining the 3-D information of the scenery by calculating image pixel deviation with triangulation principle.A entire binocular computer vision system generally contains gaining of image, pretreating, extracting of feature, stereo matching, calculating and interpolating of depth, etc. This paper mainly focuses on feature extracting, stereo matching and discrete point interpolating. Firstly, according to the stereo vision system model, the thesis sets forth the basal principle of stereo vision. Secondly, the thesis carries out detailed analysis on pretreatment methods, such as feature extracting, and checks up the algorithm by examples. At the same time, it makes a summarization to the problem of stereo matching, including the way and the quality. Finally, through the research of feature extracting and image matching, the thesis brings forward the matching method, which is grounding on features of images. With stereo obtained from the matching process, it restores the information of depth, and then various interpolation processes have been used to these discrete depth information which make a fairly good reconstruction result.