The3D Point Cloud Registration Technology Of Large Scale Shape Measurement

The3D shape measurement is an important research content of the visualmeasurement. The large dimension shape measurement generally needs3D pointcloud registration technology to get the complete morphology characteristics. So the3D point clouds registration technology has become the key link of3D shapemeasurement. In the process of the registration implementation, the splicing accuracyis an important index to measure the whole accuracy, which especially needs to payattention to. Since the large size3D shape measurement has the large scale space andthe great amount of point clouds data, the automatic data acquisition technology is animportant problem.This thesis studies a kind of3D point cloud automatic scanning acquisitiontechnology based on robot and laser vision sensor which will make full use of robotplatform’s high agility and laser vision sensor’s non-contact measuring advantage torealize the fast data acquisition within the robot work space. This thesis alsoformulates measurement system solutions. Using point clouds data by differentposition areas of the robot as the foundation, it uses the neighborhood splicing methodto splice the regional data and finally control overall splicing precision by using thebenchmark scale as the length constraint which can achieve the high efficiency largedimension3D shape measurement in the premise of ensuring the whole accuracy.The main contents of this thesis are shown as following:1. After studying and comparing the common methods and equipment of3Dshape measurement technology, this thesis presents a new3D point clouds automaticscanning acquisition method based on robot and laser vision sensor.2. Analyze and design the measuring hardware system, the system includes:robot, sensors, motorized precision rotary stage and communication control; study thecomposition of the coordinate system and structure parameter calibration; discuss thepath plan of the scanning.3. Research the ICP algorithm based on neighborhood splicing to realize thevisual data splicing. After the pretreatment which includes denoising, smoothing andsparse changing the point clouds, study how to combine the k-d tree method whichcan reduce the search time and improve the efficiency of the splicing. 4. After analyze the limitations of the ICP method, propose a new method basedon ICP combined with method of reference scale length constraint to control theaccuracy of splicing.5. Make the experiments to confirm the validity of the theoretical analysis.