Research On Segmentation Of Regular Laser Point Cloud Based On Geometric Feature

With the constantly development of national cognitive ability, Two dimensional geographic information expression has been unable to meet the demand of social development, the amount of information and the spatial topological relation which Spatial visualization model expressed shows its great advantage, make the GIS field stand on a higher level stage. The spurt progress of laser technology makes the 3D laser scanner emerge as the times require, it become a tool of modeling because the high accuracy of point cloud it can quickly obtain from the 3D scene. Compared to the laser scanning equipment, the processing of laser point cloud seems to lag behind others. The data format is simple, space correlation is not strong, unable to effectively eliminate the noise points during the collection course of data, and the point-model does not correspond to the structure shape of spatial entity, increase the difficulty for subsequent model reconstruction. Point cloud segmentation is a step of model data preprocessing, if you jump this step, modeling method turn to be more complicated,seriously affects the efficiency and quality of the model reconstruction.Based on the above background, This article has done some research on point cloud segmentation, use reasonable data indexing method to manage the discrete point data, split the regular laser point cloud based on geometric feature. The main work contents are summarized as follows:1) Analysis the spatial distribution characteristics of data, use octree partition way to organize the data, and named it in octal code to increase the point cloud space correlation, combined with K neighborhood search method to obtain the contiguous element of the target point, to calculate the space feature of data.2)Do some comparison and improvement when calculating the normal vector and curvature information. Put forward a kind of triangle normal adjustment method, we adjust the triangle vertex order to make the normal vector point to the same side, and use the weighted average of adjusted normal vector to estimate the normal vector of a point. Comparing normal distance conditional with Euclidean distance conditional,using the more reasonable fitting surface to calculate the curvature.3) Discuss and quantify the segmentation constraint conditions of regular point cloud, extract planar point by the normal vector information, according to the difference of curvature between the two principal directions, to split and recognize the cylindrical and spherical information.