Research On 3D Surface Reconstruction Based On The Image Matching Point Cloud

3D surface reconstruction technology is to restore real 3D surface shape of the object, it is widely used in computer graphics, computer animation, reverse engineering, virtual reality, and so on. How to quickly and fully automatic, low-cost constructing 3D surface model is one of the hot topics in the study of 3D surface reconstruction technology. In many 3D surface reconstruction methods, digital photogrammetry technology is also one of the important means of 3D surface reconstruction. Dense point cloud and abundant visual information can be obtained from images. 3D surface reconstruction based on image helps to improve the efficiency of 3D surface reconstruction and decrease the cost of reconstruction. Based on image matching point cloud as a data source, considering the matching source information of point cloud, the key technology of 3D surface reconstruction of the object is studied.Although image matching point cloud data can offer very rich information, but it will lead to a huge amount of point cloud data. One-time deal with large amount of data is very unrealistic, this is about to consider to partition the whole piece of point cloud data. Using the thinking of "partition and merging", firstly, 3D surface mesh should be built from a single piece of point cloud, then merging multiple 3D surface meshes to restore the shape of the original scene.Main research content of this paper is as follows:(1) The point cloud partition. This paper discusses the commonly used methods of point cloud partition, including: strip partition, grid partition, quad-tree partition and adaptive partition. In this paper, a method of point cloud partition based on adaptive partition is proposed, it maintain a certain degree of overlap between block and block in the process of point cloud partition, so that the merging of multiple 3D surface mesh.(2) 3D surface reconstruction from point cloud data. There are many kinds of 3D surface forms, in this paper, a more stable triangular mesh is selected to represent 3D surface. The commonly used methods of triangle mesh structure are discussed, including: distance field method, curved surface fitting method and Delaunay triangle subdivision. In this paper, the Delaunay triangulation subdivision method is selected to build the initial Delaunay tetrahedron. According to the visual information of tetrahedrons which come from 3D point matching source information, we get real 3D surface mesh.(3) 3D surface mesh post-processing. Preliminary reconstruction of 3D surface mesh still has some problems, it need to undertake the corresponding work some post-processing, including 3D surface mesh optimization and the merging of multiple 3D surface meshes. In the optimization of 3D surface meshes, this paper mainly has carried on the following work: removing of error triangle on the edge and redirecting of triangle normal. According to the properties of the homologous block, in this paper, the multiple 3D surface meshes adopted merging method by line per column, it put them into a large 3D surface mesh.Finally, in this paper, the point cloud at different stages of 3D surface reconstruction, respectively by three sets of data in different parts of the experiment shows the effect, and its feasibility was verified. Through two set of experiment data, we analyses the time consumption, memory consumption and problems. In addition, in order to verify the validity of the algorithm, this paper implements the Possion surface reconstruction algorithm and compares this method with Possion method. This method and Possion method from the mesh effect, time consumption and memory consumption has carried on the contrast, reflects the advantages and disadvantages of this method. 3D surface reconstruction based on image is a complex research, improving the 3D surface mesh accuracy, increasing the speed of 3D surface reconstruction and automatic texture mapping also need to be studied.