A Registration Method For Terrain Laser Scanning Data Based On Feature Combination

Terrain Laser Scanning system(TLS)acquires the 3D terrain surface information by non-contact laser ranging.TLS data acquisition has the characteristics of high density,high precision,good efficiency and flexible station setting,which plays an important role in 3D city reconstruction,engineering surveying,protection of cultural relics,natural forest and field terrain analysis and so on.The laser scanners are usually fixed as access to data,because of perspective of the occlusion problem,the complete 3D point clouds are often obtained by multi station scanning.Therefore,the registration of multi station point clouds has an important position in laser point cloud data processing.TLS point cloud data is usually in the scanner relative coordinates,in which the relative position and attitude is unknown,the problem of TLS registration is essentially the problem of rigid transformation parameters calculating among different stations.In practical,there are several manual registration solutions,such as setting up markers or specifying the matching points manually,but these methods are often time-consuming and laborious,which seriously affect the efficiency of data processing.Therefore,finding an automatic method of laser point cloud registration is of great importance in the laser point cloud data processing.There are two kinds of main methods for automatic registration of laser point cloud.One is the matching method based on feature point,using the feature descriptors and some matching strategies,calculating the transformation parameters from matching of corresponding points automatically,but the performance of feature descriptors is not stable in the condition of drastic changing of point density and shielding case which is common in TLS data,the higher overlapping degree is needed;another method is the matching method based on geometric features,which extracts geometric elements from point clouds by different types of algorithm at first,calculates the registration parameters using the mathematical model of geometric elements,the effect is relatively stable but the applications is not extensive.This paper puts forward the idea of multi-class geometry elements combination,which extracts geometric feature elements through certain algorithms as registration primitives,solves the rigid transformation by traversing the combination between two station features,and find the best transformation parameters through the fitment voting method in the end to achieve rough registration.Based on rough registration,this method automatically constructs station distribution graph,calculates the optimal transformation parameter sets in the sense of least squares using graph optimization method based on the geometric features.The multi-station registration scheme in this paper use three kinds of geometric primitives(patch elements,pole elements and edge lines)for clues,achieves the complete procedures from registration between two stations to multi-station point cloud output,which can solve the registration problem of big scene with large amount of data with a high degree of automation and robustness.Experimental results show that the method is stable and reliable,and has good computational efficiency and accuracy.