Research On 3D Reconstruction Technology Based On Visual Images

With the development of computer technology and 3D reconstruction technology,3D reconstruction based on visual image has become one of the hot topics in computer graphics.3D reconstruction can be divided into two categories,one is 3D reconstruction based on single image and the other is 3D reconstruction based on video sequence.As an indirect three-dimensional modeling technique,this method is widely used in the fields of film making,cultural relic restoration,mechanical manufacture and medical health.This paper introduces the shape from shading(SFS)method and binocular stereo vision method,which can use single picture or two pictures effectively reconstruct the 3D shape of the object.In this paper,the 3D reconstruction method based on single picture is first introduced,and the advantages and disadvantages of the SFS method and the deformation model method are analyzed.Because of the deformation model method requires a priori database information,which greatly restricts its application.The SFS method can reconstruct the 3D shape of objects based on single image,which is simple and convenient to use.At the same time,the measuring principle and matching method of binocular stereo vision are introduced,and the advantages and disadvantages of different matching methods and the scope of application are analyzed.The SFS method is an ill posed problem;therefore we need to increase the constraint conditions,such as general brightness constraint,smoothness constraint,intensity gradient constraint and unit normal vector constraint,and based on this basis it can be divided into four classical algorithms:minimization method,evolution method,local method,linearization method.During the binocular stereo vision reconstruction process,the choose of matching unit,similarity measure,search space and search strategy will affect the final reconstruction accuracy.In the selection of the final matching method,the matching method based on the region correlation or the matching method based on feature can be selected.In order to solve the problem that the Lambertian reflection model is difficult to describe the real reflection in the traditional SFS method.A hybrid reflection model integrating Lambertian reflection model and Phone reflection model is established,and the corresponding diffuse reflection and specular reflection can be reflected in the hybrid reflection model,which improves the accuracy of 3D reconstruction.In the process of binocular stereo vision,the corner extraction process is optimized,and the optimized Scale Invariant Feature Transform(SIFT)algorithm is proposed.The corner selection process is added before extracting corner points which reduces the search area.Compared with the traditional SIFT algorithm,the accuracy increases by 40%.At the same time,in order to solve the problem of mismatch in the process of feature matching,the feature point pairs obtained from the initial matching are screened by constraint conditions.Thus,the influence of the wrong feature points on the reconstruction results is avoided.This experiment uses VS2013 as a software development platform and object-oriented C++ language as a development language.According to the principle of interface design and the requirement of 3D reconstruction,the 3D reconstruction interface based on SFS and binocular stereo vision is designed by MFC and OpenCV.