Research on Multi-View 3D Point Cloud Registration Method Based on Improved Shape Generation

Three-dimensional laser scanner is widely used in the fields of surface retest, model reconstruction, virtual simulation scene acquisition, etc. For the data processing of three-dimensional laser scanning point cloud, there have been a lot of studies on the automatic registration of paired three-dimensional point cloud, but there are relatively few automatic registration of multi-view three-dimensional point cloud. Due to the larger number of parameters to be solved for the registration of multi-view point cloud, As well as the accumulation of errors, multi-view point cloud registration is more challenging than two-view point cloud registration. In order to solve the cumulative error problem caused by multi-objective registration, it is urgent to explore an optimization algorithm, which aims to optimize the registration calculation process of scanning point cloud data and improve the registration efficiency and accuracy of high-point cloud. This paper introduces the principle of multi-objective point cloud continuous registration, and proposes a multi-view three-dimensional point cloud registration based on shape generation, which is used to solve the problem of multi-view point cloud registration with topological relations. In view of the shortcomings of traditional shape generation algorithms in rough registration of multi-view point clouds, improvement measures are given: (1) Select the point cloud with the largest size as the seed shape point cloud to reduce the registration failure probability; (2) The largest point cloud is selected successively according to the size of the point cloud for registration, so that the updated seed shape point cloud has more common information with the subsequent point cloud, and the probability of successful registration is increased. Finally, the improved shape growth algorithm is used to solve the registration problem of point clouds with tree topology relationship. The experimental results show that the improved shape growth algorithm reduces the registration error of point cloud with tree topological relation by 5.5% compared with the direct registration error, and has an improved effect on the reduction of continuous registration error of multi-target point cloud.