Study On Deformation Monitoring Based On LiDAR Point Clouds

With the maturation of hardware equipment for the terrestrial laser 3D imaging system and the continuous iteration of data processing software functions,terrestrial 3D laser scanners are increasingly being used in space measurement,model reconstruction,and structural health monitoring.Based on the characteristics of high density,high precision,and the large amount of information contained in point clouds,point cloud preprocessing and application have become popular in recent years.In this paper,a laser scanner was used to collect field experiment data for multiple different scenes.The method including preprocessing of point cloud data,threedimensional model reconstruction,and point cloud data were used to analyze the minor deformation of target,and algorithms improvement and analysis methods was raised correspondingly.This article focuses on detailed and in-depth research on the following issues:The spatial geometric characteristics and spectral characteristics of point cloud data are analyzed in detail.K-d tree is used to divide the disordered point cloud,and its topological relationship is established to improve the query efficiency.Echo reflection intensity,RGB and other characteristics of the point cloud data were analyzed.What's more,colored point cloud model was constructed too.On this basis,the common algorithm of point cloud segmentation is studied and an improved method is proposed to improve the segmentation accuracy.Based on common algorithms,this paper combines the color characteristics of point cloud and the echo reflection intensity to propose the region generation segmentation algorithm based on echo intensity.The geometrical feature is used as the primary judgement of growth conditions,and the echo intensity is used as a supplementary judgement to improve the accuracy of point cloud segmentation.The feasibility of the algorithm is verified by experiments.When it comes to point cloud registration.Intensive research and analysis were conducted on the nearest point iteration method and normal distribution transformation method.In this paper,a multi-step registration strategy was proposed whose name is “NDT coarse matching for the same name point and ICP accurate registration of dual-site data with multiple variation thresholds”.The strategy quickly finds a closer initial position transformation matrix.By gradually and accurately aligned,the registration efficiency increases by an average of 50% and registration accuracy reaches the order of 0.001 m.For the reconstruction of 3D point cloud model,the process of Poisson reconstruction and Delaunay triangulation reconstruction is studied.Greedy projection triangulation method is used to reconstruct the model.After analysis,it is found that there are many holes on the surface of the model.In order to improve the accuracy of this model,greedy algorithm is sectionally iterated and the coordinates of the missing parts are predicted by the SVM library of the PCL point cloud library.The experimental data is modeled and a smooth three-dimensional model is obtained.The standard deviation of the three-dimension model is 0.2345 mm.Based on the above researches and point cloud processing steps,reservoir dam in Yingde City,Guangdong Province was selected as the experimental observation target.After designing the field acquisition experiment,the above algorithm flow was used to perform point cloud data processing and the establishment of a three-dimensional model.The model of double-station data was quantitatively analyzed from view of point-to-point and model difference to get the deformation characteristics and good experimental results were obtained.