Research On The Method For 3D Reconstruction Based On Stereo Vision

The research purpose of 3D reconstruction based on the stereo vision is toreconstruct the information of 3D Scene by using 2-d image information. Because ofthe strong pointing of low cost, simple operation and high realism, this method hadbecome into hot topic in the field of computer graphics and computer vision. Moreover,it had been used widely in object recognition, robot navigation, virtual reality andindustrial test. In 3D reconstruction, stereo matching is the hardest and most importantstep, because it concerns much. This paper research system image matching and 3Dreconstruction, and do the related test.This paper instruction and research the following issues.1. This paper introduces three kinds of ordinate system in the computer vision field andcamera model. Moreover it exposits the basic concept of epipolar geometry,fundamental matrix and homography matrix. Meanwhile, the research content of stereovision is introduced in this paper.2. This paper introduces popular feature point extracting algorithm and matchingalgorithm, and make the related experimental analysis. This paper proposed an imagematching method based on affine transformation. Moreover this paper analyzes thenecessity of affine transformation in some cases and presented the experimental dataand result. This method avoids the shortage of low precision by using polar constraintsonly and rare matching points. Experiments show that this method is of high precisionand speed.3. This paper proposed a matching method basing on homography constrain. In thebasis of epipolar constraint relationship, the homography constraint matching is donethough solving high-precision homograhpy matrix.. It is possible that single epipolarconstraint can't determine uniquely all matches, because of lighting conditions, noise,distortion and the autologous character of camera and other factors. This method hasmore obvious advantage in matching. Experiment proved that this method is with thecharacter of high-speed and high-precision.4. This paper introduces the solving methods of fundamental matrix, roate matrix andtranslational matrix, and some key technology of 3d reconstruction such as Europeanreconstruction, gridded 3D point cloud, texture mapping etc.. Further this paperdescribes the process of 3D reconstruction systemly and illustrates the function of each module. Finally this paper proves this system is workable by experiment and analyzesthe experimental precision.