On Study Of Reconstruction From Large-scale Architectural Point Cloud

Obtaining accurate three-dimensional model of the actual object is a hot spot in computer graphics research, and is widely used in reverse engineer, medical image processing, virtual reality and visual inspection. With the development of the point cloud data acquiring technology, some special equipment were invented, efficiency of modeling architecture is greatly improved. Extracting point cloud and reconstruction of historic buildings, not only to achieve virtual roaming, but also can be applied to the repair of them after damage. The actual architecture models can be used for simulation of war, fire or terrorist attacks and other special circumstances, in favor of analyzing the present conditions and dealing with them quickly and accurately.This paper researches the main point cloud processing technology. For the point cloud of architecture is dense and has a large volume of data, use Voxel Grid to down sample is necessary. Calculate the number of points around each point to determine the outliers. Compared reconstruction based on implicit functions with reconstruction based on topology, we analyze their characteristics and determine their scopes.Building point cloud often accompanies with plane, using the sampling consistency algorithm(RANSAC) to extract them. Euclidean clustering analysis can extract the point cloud of individual object on the planes, which can be used for reconstrction. This method can automatically get point cloud of objects used for reconstruction on the planes.Since getting triangular meshes from point clouds are often accompanied by holes under existing conditions, so repaire them is necessary. based on implicit function to repair the model can obtain a smooth result, an automatic algorithm that could repair the holes based on RBF(Radial Basis Function) is presented in this paper. By calculating the correlation of the holes on the main body of mesh and isolated surfaces to determine the points used to calculate RBF. This method is divided into two steps: Firstly, use the minimum weight triangulation method to fill the holes on the main body of the mesh. Then, the vertex obtained by subdivision is adjusted to the surface expressed by RBF.The new triangles and the original triangles are smoothly connected. The results show that, this repair method can automatically repair all models and get more accurate models.