Linear-Feature-Constrained Registration of LiDAR Point Cloud via Quaternion

Considering the large amount of computation & low accuracy of extracted point-like features are the two main disadvantages of traditional point-to-point based registration methods which is designed for LiDAR point cloud,and the accuracy of registration results is seriously decreased by the linearization procedure of traditional 7-parameter based transformation approaches,a new registration approach is designed to overcome above disadvantages,which selects linear features as registration primitives,and uses quaternion to represent rotation matrix.Similarity measure of the linear-feature-constrained 3D transformation procedure is presented,and the formulation of registration procedure is exactly deduced.Besides,the detailed procedure of how to calculate rotation,translation & scale is also presented.Experiments show that the presented approach is efficient & effective.More importantly,by using quaternion to represent rotation matrix,the new presented approach avoids the decrease of accuracy,meanwhile,due to the characteristic of quaternion,it also needs few calculation resources compared to traditional registration methods.