No Deformation Texture Mapping For The Three-dimensional Laser Scanning Technology Research

3D laser scaning and modeling has been developing fast in recent years since 1980s. This technology has been used widely in many fields. But most of the algorithms concerning 3D reconstruction are designed for surface reconstruction based on unorganized and large-scale pointcloud, which focus on how to reconstruct models rapidly and accurately, but neglect texture features, so can not satisfy the need of highly realistic geometric models for some specific applications, such as cultural preservation, cite monitoring, digital cinematography, etc. The above-mentioned issue is studied in this paper, and the main contributions are as follows:(1) The difficulty in constructing a model with realistic and undistoryted texture is discussed after a review about the 3D laser scanner in our laboratory and the modeling processure. Then undistorted texture mapping is presented and the solution is proposed in order to produce models with visual reality, which includes the study on registration of 2D texture and 3D points and reconstruction of the triangulated model.(2) The basic concept of registration of 2D texture and 3D points is given, and a variety of existing registration methods are summarized and analyzed. Then a novel registration algorithm is proposed, in which the position and orientation of the camera relative to the 3D scanner are computed according to the properties of vanishing point. This algorithm provides technical assurance for the flexibility of data acquisition.(3) The triangulated surface models constructed based on 3D pointcloud usually need to be simplified to facilitate the storage and transmission, so such a problem often arises: a triangle after surface simplification can be so large that its three vetex correspond to pixels in two texture images, and thus can not be textured directly. An algorithm of model reconstruction based on texture image border is proposed, which provides the posibility of texture mapping for largely simplified models.The above algorithms are all realized on the platform VC++ and OpenGL, and also tested with the data acquired by the laser scanner in our laboratory. Experimental results prove their validity.