Research On The Multi-Feature-based Registration Of Terrestrial LiDAR Point Clouds

Light detection and Ranging(LiDAR)has been widely used in protection of cultural relics,surveying and mapping,civil engineering and other fields because of its excelent characteristics such as non-contact,high efficiency and high precision.Because of the high spatial complexity of the target,it is usually necessary to deploy stations along different directions and scan the target.In order to obtain the complete point cloud data on the surface of the object,registration is needed to realize the splicing and fusion of LiDAR point clouds from different stations.Because of this,the quality of registration results has a direct impact on the effect of point cloud post-processing,such as three-dimensional modeling.According to the different feature extracted from point cloud,the registration algorithms divided into three categories: point-feature-constrained,linear-featureconstrained and planar-feature-constrained registration algorithms.Most algorithms choose one of the three features as a constraint.However,in cities,there may be insufficient constraints in simply selecting a class of features as registration primitives.As a basic model for describing spatial similarity transformation,Bursa model has been widely used in LiDAR point cloud registration.According to the different mathematical descriptive tools chosen for spatial similarity transformation model,the registration algorithms are divided into two categories: vector algebra-based and Clifford algebrabased.At present,Most of them choose vector algebra as the basic tool of mathematical description.They describe rotation by angle or matrix,and then solve the registration parameters.However,such methods have the following shortcomings: 1)the description of spatial rotation transformation is complex;2)the initial value of the parameters is dependent.Based on the above analysis,aiming at the characteristics of serious occlusion in urban construction-intensive areas,registration model based on point,line and plane constraints is developed.The specific work is as follows:1)A Dual Quaternion based,Point Feature/ Linear Feature/ Planar Feature constrained Registration Approach for Terrestrial LiDAR Point CloudsTaking Bursa model as the basic registration model,dual quaternion as the description tool of spatial rotation,three-dimensional coordinates,Plücker coordinates and six parameters(normal vectors and passing points)were selected to describe different features respectively.Based on the precondition of coincidence between features,the corresponding objective function was constructed.By linearizing the objective function,the registration parameters of LiDAR point clouds with different kinds of feature constraints were solved by iterative calculation.2)A Dual Quaternion-based,Multi-Feature-based Registration Approach for Terrestrial LiDAR Point CloudsWith the constraints of point feature,linear feature and planar feature,the initial iteration value of model parameters was solved by rough registration using the conjugate features.Then,based on point feature constraints,the corresponding objective function was constructed by adding the constraints of line feature coincidence and normal vector parallelism.By taking the transformation parameters obtained from rough registration as the initial values,the registration parameters were solved iteratively based on least squares method.Above all,the comprehensive utilization of the constraints between LiDAR point clouds of adjacent stations is realized,which provides sufficient constraints for the high-precision registration of LiDAR point clouds and effectively guarantees the accuracy of the registration results of LiDAR point clouds of adjacent stations.