Research On 3d Point Cloud Automatic Registration Algorithm

Lidar scanning technology is becoming more and more mature and has been applied in many fields.In reverse engineering,the protection of cultural relics,surveying and mapping and geographic information system,automatic driving,movie and game industry,and other fields have a 3d laser scanning technology,but the precision of the laser radar equipment analyte with bright and clean degree on the surface of the object,the influence of such factors as the laser point cloud is uneven sampling,the characteristics of the loss,and a lot of noise and other issues,so in order to improve the accuracy of point cloud late 3 d modeling,need to deal with point cloud,the existing problems,this paper point cloud of point cloud filtering,point cloud simplification,feature extraction and the point cloud registration problems for research.The research contents and results of this paper are as follows:(1)Aiming at the problem of noise point cloud,this paper studies the radius of the filter,statistics filter,and bilateral filtering the three filtering methods,using the radius of the filter can better monitoring of outliers,the point cloud radius of the filter in the first place,at the same time guarantee the point cloud using statistical filtering characteristics on the basis of bilateral filtering method is used to ensure that the point cloud filter has the characteristics of the fair sex,to ensure the effect of the point cloud filtering.Experiments show that the proposed algorithm can simplify the point cloud quickly and save the local features of the point cloud.(2)In view of the large amount of data redundancy in point cloud,this paper studies two methods of point cloud simplification of voxel grid and curvature fitting simplification.Voxel grid simplifies point cloud computing speed,but the point cloud features cannot be well preserved,and the point cloud feature extraction based on curvature is slow.In this paper,the region with a large change in point cloud curvature is proposed,and a small voxel grid method is adopted to preserve the features of the region with a large change in point cloud curvature,and more redundant points are removed in the region with a small curvature,so as to ensure the simplification of point cloud and the preservation of local features of point cloud.Experiments show that the proposed algorithm can simplify the point cloud quickly and save the local features of the point cloud.(3)Aiming at the problem that ICP of the classical point cloud algorithm falls into the local optimum and the iteration speed is slow,this paper uses the method of m-estimation to carry out selection iteration to eliminate the error-matched feature point pairs in the point cloud,so as to increase the accuracy of point pair matching in the point cloud feature,reduce the number of point cloud iteration and improve the precision of the higher cloud.Experiments show that the proposed algorithm improves the accuracy of point cloud registration and reduces the time of point cloud registration compared with the traditional near point iterative algorithm.