Research On Registration Technology Based On 3D Feature Point Cloud In Indoor Environment

With the wide application of 3D scene reconstruction technology in the fields of robot vision,robot perception,virtual reality,and reverse engineering,it is difficult for the existing collection equipment to accurately perform accurate 3D scenes at one time due to the field of view and object occlusion and the performance of the equipment itself.Complete model reconstruction,so it is necessary to collect multiple sets of point cloud data from different perspectives at multiple sites for registration.The most classic point cloud registration algorithm is the iterative closest point registration algorithm ICP(Iterative Closest Point).The advantage of the ICP algorithm is that the registration accuracy is high,but this algorithm has high requirements for the initial pose of the registered point cloud.The effect of point cloud registration with large initial pose deviation is not good,and it is easy to fall into a local optimal solution.In order to reduce the deviation provided by the initial pose of the point cloud,the current mainstream 3D point cloud registration algorithm divides the registration into two stages: coarse point cloud registration for the initial pose and point cloud fine registration for the matching points.The main research contents of this article include:(1)An adaptive voxel grid filtering algorithm is proposed to filter the original data point cloud,and the side length of the voxel grid cube can be adaptively modified according to the amount of original three-dimensional point cloud data to be sampled.Downsample the number of point clouds to a preset number.This method can effectively reduce the data size of the 3D point cloud under the premise of ensuring the stability of the normal vector and curvature of the initial 3D point cloud data,thus reducing the calculation time of the point cloud registration and improving the overall point cloud registration algorithm effectiveness.(2)The sampling consistency quadratic distance registration algorithm SAC-QA(Sampling Consistency Quadratic Distance Aligment)is proposed,and the secondorder function of distance is introduced as the penalty function in the FPFH,which further reduces the distance between neighbors.The weight of the domain point,while increasing the weight of the nearby neighborhood point.Then,the traditional ICP registration algorithm is used for fine registration,and the final registration result is obtained.(3)Designed and implemented a point cloud registration system based on 3D features.The system realized the import of 3D point cloud public data sets such as Bunny,Happy,Dragon,Armadillo,etc.of Stanford University,point cloud preprocessing,and initial Registration,precise registration and display of registration results,this is a more complete registration system.Experiments using the Bunny point cloud data public dataset of Stanford University show that compared with the improved SAC-IA point cloud registration algorithm,the algorithm in this paper improves the 3D point cloud registration accuracy by 54.65%,and is more efficient in registration.An increase of 39.39%.In addition,using Stanford University’s open data sets of 3D point clouds such as Happy,Dragon,Armadillo,etc.,the registration time can basically be maintained within 10 seconds,which verifies the effectiveness of the algorithm in this paper.