Research On Registration Method For Point Cloud Obtained By Structured-light Sensors

With the upgrading of point cloud access device and the rapidly development of 3D point cloud processing technology,3D point cloud processing technology with the potential of the prairie fire was widely applied to the protection of cultural relics and archaeological studies,digital entertainment,military and engineering geological mapping.To get a full three dimensional digital model,we must measure to get point cloud data for subsequent processing.Point cloud registration technology as a necessary link from multiple measurement datum to complete model was a priority for building 3D digital model.The paper makes a deep research on previous work and put forward two kinds of solution for point cloud registration.First of all,a brief introduction about research background,implementation prospect and research situation at home and abroad of the subject research was made.Through analyzing research situation of point cloud registration technology solutions,registration algorithm was divided into three categories: the registration method based on feature matching relation,registration method based on the statistical rule,iterative least error registration method.Paper introduces the basic concepts of point cloud registration,and lay the foundation for the subsequent discussion papers.Secondly,the paper studies and summarizes details of the previous work,then,get a large amount of design inspiration and point cloud registration algorithm based on FPFH(Fast Point Feature Histogram)was designed and implemented.Through the ISS algorithm,we can get key points of point cloud data;The FPFH features of every key point can be the foundation of the follow-up point cloud processing;The random sampling consistency algorithm will be used to remove error corresponding relationship preliminary determined between the two measurement point cloud data;Then covariance matrix built by the corresponding relations decomposed by singular value decomposition to obtain the rigid body transformation matrix.This kind of point cloud registration method has higher registration precision and speed,can meet part of the Engineering requirements.Thirdly,further studying of previous work,a registration algorithm of point clouds using multi-scale normal feature was proposed.The algorithm's innovation is mainly reflected in three aspects: key point selection based on multi-scale curvatures,the point feature descriptor based on the multi-scale normal vectors feature and corresponding relation based on minimum and second distance clustering sorting.Key points in the process of registration were obtained through the multi-scale curvatures get by the principal component analysis(PCA)and the objective function;Deviation between multi-scale normal vectors of a key point was made to Key points feature descriptor;Choice is critical,the corresponding relationships which was very important were Preliminary get by minimum and second distance ratio between two features.After using random sampling algorithm and clustering method,accurate corresponding relations were get.Corresponding relationship of the optimized can be Construct covariance matrix.We can get parameter matrix by singular value decomposition between two point clouds.Finally,enforceability and the engineering practicality of two methods were verified by using a variety of point cloud data.With the analyze about speed and accuracy,we can get conclusion that point clouds registration algorithm using multi-scale normal feature have fast calculation speed,small computational complexity and high Registration precision.This method need not second registration and has Anti-noise ability,which has practical value to the project.