| LIDAR point cloud data is an important data source for measuring forest trees and retrieving individual tree parameters.It is a research hotspot to extract forest biomass in a wide range and with high precision through laser point cloud data and analyze the distribution of forest carbon reserves.However,there are defects in the point cloud data obtained by different instruments.The Terrestrial Laser Scanning(TLS)can better obtain the understory structure,but the scanning in the upper canopy is limited,while the Airborne Laser Scanning(ALS)can widely obtain the upper canopy structure,but cannot accurately scan the information in the forest.The registration of the two is combined to realize the complementary perspective of data,which is helpful to improve the point cloud data and improve the inversion accuracy of forest parameters.This paper focuses on the in-depth research on the forest point cloud data registration technology of TLS and ULS and puts forward a registration method from the height point of the tree.The tree height points are obtained from a digital surface model(DSM),which contains isosurface points of the forest structures.Rotation and translation matrices are then calculated through singular value decomposition(SVD),and rough registration is completed.Finally,fine registration is achieved through nearest-neighbor iterative SVD,and we verify the registration accuracy and compare it with multiple methods.The main research contents and results of this paper are as follows:(1)In this paper,the point is transformed into a disk to realize the transformation from point cloud to image,which can effectively reduce the white line pits in the process of rasterization and fill in blank information.Then,by calculating the sample plot center point,the ULS point cloud data is transformed from global coordinates to local coordinates,and the superposition of ULS-TLS point cloud data is preliminarily realized.Then,from the perspective of the tree high point,TR matching is proposed.The method includes coarse registration and fine registration.The former searches the tree high points of ULS and TLS images through the search window,and carries out similarity recognition and registration.Then,the highest point after rough registration is iteratively re-registered to improve the accuracy.The average distance between the nearest points of the two-point clusters after registration is calculated as the registration accuracy.The coarse registration accuracy is between 0.38-0.83 m,the fine registration accuracy is between 0.18-0.69 m,and the total time is no more than 30 s.The fine registration can effectively improve the coarse registration results by 55%.The TR method has achieved good accuracy and time benefit as a whole,and good results are obtained in the high mountain forest with a slope of 2-21 °.(2)In this paper,ICP,CPD and NDT are used to register ULS-TLS point cloud data in two states before and after rough registration,but the effect is better.Before the two data rough registration,ICP is used for registration,the accuracy is 0.85-34.04 m,NDT is 1.0-33.11 m,CPD is 19.26-25.83 m,the overall accuracy is 0.85-34.86 m,and the range of accuracy changes Large,all three are affected by the initial position,and can not get better results.After the coarse registration of the ULS-TLS point cloud is realized by the TR algorithm,the position of ULS-TLS point cloud data is adjusted.The results of fine registration using three methods are not ideal.The precision of ICP is 0.87-34.86 m,that of NDT is 0.62-8.25 m,that of CPD is 17.93-23.87 m,and that of the overall precision is 0.62-24.57 m.The three methods do not improve the precision of coarse registration,On the contrary,randomness and precision change jump appear,and their registration accuracy is far lower than that of the TR fine registration method.(3)Taking sample plot 1,sample plot 2 and sample plot 3 as examples,this paper divides the search window into five sizes: 2m * 2m,3m * 3m,4m * 4m,5m * 5m and6 m * 6m,and analyzes the impact of the change of the size of the search tree high point window on the TR registration algorithm.It is found that with the change of the search window from small to large,the searched tree high points are decreasing in both ULS and TLS point cloud data,and are far from reaching the sample size The number of single trees in the land itself.When the search window is 2-5m,the accuracy of final registration changes stably,fluctuates little,and the registration accuracy is high.However,when the search window exceeds 3m,the accuracy of sample plot 3 is too low,which is greatly affected.When the search window is 3m,the registration accuracy of the three sample plots is the best.In addition,this paper also discusses the influence of TLS point cloud data density on registration accuracy.Dilute the TLS data nine times equidistant until its distance density is similar to the ULS data,and explore the impact of density transformation on the registration accuracy,in which the window for searching the high point of the tree is 3m * 3m.The results show that TR is less affected by point cloud density.During the change of TLS data density,the change range of registration accuracy is small and tends to be flat.The proposal of the TR algorithm can effectively solve the shortcomings in the current research field,quickly and accurately realize the registration of forest space location cloud data,improve time efficiency and reduce labor cost.It has good application value. |
PDF Full Text Request