Point Cloud Registration Algorithm Based On Multi-Constraint Octree And Multiple Features

As the first step of point cloud reconstruction,the rapid and accurate implementation of the process of 3D point cloud registration affects the speed and effect of reconstruction.Due to the impenetrability of the object and other factors,it often needs different angles and many times to collect 3D point data by 3D laser scanner.It leads that the spatial coordinate system of acquired 3D point cloud data which comes from different batch is inconsistent and must be adjusted by registration.The registration process is to find the rigid transformation matrix which can convert the registration point cloud to the target point cloud.With the development of laser scaning technology,3D laser scanner can collect a large number of point cloud data in a short time.This proposes a higher request for point cloud reconstruction in speed and precision.Therefore,optimizing 3D point cloud registration process has became one of the important topics in the research of 3D point cloud technology.To optimize the process of 3D point cloud registration in speed and precision,this paper proposed a point cloud registration algorithm based on multi-constrained octree hierarchical fitting and multiple feature matching after thoroughly studied 3D point cloud registration algorithm,3D point cloud segmentationalgorithmandsurfacefittingalgorithm.Firstly,the multi-constraint is added to imporve the adaptive octree to reduce the surface complexity of the segmentation child node.Then according to the segmentation characteristics,the hierarchical strategy based on moving least squares（MLS）is adopted to reduce the fitting time.And then the multi-features of the points are calculated by the fit-surface,after that the multi-feature similarity is proposed to improve the accuracy of the matching pair.Finally,changing the iterative termination condition of the multi-resolution iterative closest point（ICP）to complete the registration.The main research work of this paper includes reviewing books and papers that related to 3D poing cloud registration to understand the development trend of research status at home and abroad and know hotspots and difficulties in this field.We analyzed the classic methods and related research results of the subject,combine with the data structure of 3D point cloud storage,features of 3D points and registration methods to improve the accuracy and speed of 3D point cloud registration.Based on the above research work,the innovations of this paper are as follows:1.A hierarchical MLS fitting based on multi-constrained octree is proposed.On the one hand,the adaptive octree algorithm is improved with multi-constrained added to reduce the number of segmentation and the complexity of child node.On the other hand,according to the characteristics of the surface in child node,hierarchical MLS fitting strategy is proposed to fit child node.When the MLS fitting function is establishing,choose different basis function based on the standard deviation normal vector in child node to reduce computation.This algorithm can maintain or reduce the time consume while improving the accuracy of the fitting.2.A point cloud registration algorithm based on multiple feature matching is proposed.It uses the fitting surface which is the results of previous step to calculate multiple features of point,including the Gaussian curvature K,the average curvature H,the principal curvatures k₁ and k₂,the distance L from the center of gravity,and the like.This method has two steps.One is to propose multiple-feature similarities to establish matching point pairs,which can reduce mismatch and improve the accuracy of registration.The second is to improve the multi-resolution ICP algorithm.The difference between the two consecutive mean square errors is used to instead of the minimum threshold of the mean square error as the iterative stop condition,which can effectively avoid useless iterations and reduce time consumption of the algorithm.