Texture Image Processing Techniques For 3d Laser Scanning

In recent years,3D laser scanning, as an important means of entities digitalization and 3D reconstruction, is gradually applied to reverse engineering, archaeological, virtual reality, ground-based observations and many other areas, which further promotes the development of 3D modeling techniques. But at present, the researchers focus more on how to fit point cloud data accurately with surface model, but ignore the model's texture features, making the surface properties and characteristics of the 3D model poorly reflected and visual realism badly achieved. Therefore, in the research of 3D reconstruction based on 3D laser scanning technology, the problems in texture need to be solved, so that the reconstructed model is not only accurate in size but also realistic in appearance.In this paper, we deal with the problems in texture images from two aspects: reconstruction of texture details and shadow removal, which are done based on the original images captured by camera and the spatial data of the model. This processing can improve visual realism of the textured models, and thus help to widen the application of 3D laser scanning technology. The main work includes:(1) The functions and principles of the 3D laser scanner and digital camera in our laboratory are analyzed, and the problems in texture images are proposed:the limited dynamic range of texture image and shadows;(2) The synthesis of high dynamic range (HDR) texture images is discussed. For scenes with large brightness range, any single image obtained by conventional cameras inevitably loses some information. But the texture images, synthesized using HDR processing technology, provide richer details and give texture-mapped 3D model a realistic appearance;(3) The shadow detection and removal in texture images is researched. Relighting algorithm is studied to deal with the problem in texture shadow, and then an improved shadow removal algorithm is presented. The usage of 3D model data makes the algorithm more reliable than those which are based on only image data.The above algorithms are realized with either MATLAB or VC++and OpenGL, and tested with the data collected from our campus. The results show the improvement in visual realism of 3D models and prove the feasibility and effectiveness of the algorithms.