Segmentation, Reconstruction, And Dimension Measurement Of Point Cloud Data Of 3D Laser Measurement System

Because of laser's resistance to harsh environment like high temperature and strong vibration, laser radar has been being widely used for non-contact measurement in various industries. A good example would be that 3D measurement system consisted of 2D laser radar and servo motor is used to measure hot heavy forgings' dimensions. Measurement system scans the forging in different planes, and massive data is obtained in a short time. The post-processing of these data directly determines whether the dimensional measurement can be realized, and whether the degree of precision meets the requirement. Hence, researches on methodology and efficiency of post-processing seem to be significant.Based on massive point cloud data obtained from 3D measurement system consisted of LMS 100 laser radar, adopted to the base study"Research on new method about on-site 3D precision measurement for large scale forging object with high temperature"(National Natural Science Foundation,No.50805094), this thesis is focusing on the following researches:1. Segmentation. The perspective of laser radar ranges from -45°to 225°. That is to say the forging just occupies a little part while most of the data does not belong to the forging at all. It is called "irrelevant data". In order to facilitate follow-up work, the forging data must be separated from the irrelevant one. In this thesis, the characteristic that points distribute in both horizontal and vertical layers is utilized to separate the point cloud in various 2D planes, which leads to a favorable separating result.2. Model reconstruction and quality assessment. Reconstructing forging's model is a vital step to achieve dimensional measurement. In the thesis, based on summarizing previous researches, a simpler and faster method for direct Delaunay triangulation is proposed. Moreover, according to error ellipsoid theory of spatial point, error ellipsoid model of triangle's normal is inferred. Then, error ellipsoids of forging's reconstructed model are calculated and showed in figures.3. Feature extraction, dimensional measurement and feature point tracking. Points'normals can be computed from normals of their surrounding triangles. Then, geometrical feature points can be extracted according to differences of local normals. In order to fit feature points, a plane is fitted firstly, onto which the feature points are projected. Then, these projected points are fitted to a curve, and dimensions can be computed simultaneously. At last, the minimum-norm-of-the-similarity- matrix method is presented to match the feature points, which results in tracking a certain feature point.This thesis is focused on massive point cloud data obtained from 3D laser measurement system. Based on sufficiently analyzing characteristics of data and summarizing prior work, a series of new and optimized methods for post-processing are proposed. These methods are meaningful to both engineering application and theoretical research. The thesis also has reference value to processing of point cloud data obtained from others ways, not just limited to laser radar measurement system.