Research On Scattered 3D Point Cloud Data Pre-processing Denoising And Registration

The rise of 3D scanning technology has benefited from the rapid development of electronic information technology in the past half-century.On this basis,3D scanning equipment has also been widely used in various fields.Due to the natural defects and errors of the system,the results are inevitable.There will be noise.In this regard,we must perform pre-processing operations such as noise reduction on the data to ensure the validity of the collected point cloud data,and to a large extent,the original object itself should have smoothness and smoothness characteristics.Denoising is a research hotspot in point cloud data preprocessing technology.After theoretical research on the development and research status of 3D LST and domestic and foreign,this thesis has done the following work on the design and implementation of scattered point cloud data denoising algorithm: Firstly,based on the study of spatial hierarchy,the disorder The scattered point cloud data of the structure is selected based on the k-d tree search method to construct the topology structure and realize k neighbourhood search.The concept of point cloud differential geometry information and the whole process of point cloud pre-processing work are introduced,and the data pre-processing is analysed,which the key technical links in the process.In order to remove discrete large-scale noise,this thesis proposes a VSCD(variable scale combined noise reduction)algorithm: firstly,calculate the density information of each sampling point,filter out a kind of noise points;then use clustering based on region growing the segmentation algorithm filters out the second type of noise data that is fragmented away from the main point cloud.The experimental results show that by setting a reasonable threshold and detecting the two kinds of noise step by step,the large-scale noise in the point cloud data can be expertly filtered out.In the surface and edge regions of the data model,there are usually stable noise features.In this thesis,the bilateral filter was improved: 1)Based on bilateral filtering,a Gaussian curvature kernel function is added to make the sampling point When calculating the filter factor,it is constrained by three weighting factors: distance,average and curvature.2)Based on the Gaussian curvature weight,the distance threshold of the RANSAC fitting plane is used as the sufficient discriminant conditionof the curvature kernel function,so that the curvature function is active with noise.Weak adaptive adjustment.Experiments show that by continuously smoothing the coordinates of the noise position,it can effectively reduce the interference of sharp noise at the edge position on the user data.Finally,this thesis uses the ground-based 3D laser scanning instrument to collect point cloud data from multiple stations and designs and implements the development of the point cloud fast registration system program.By using the algorithm proposed in this thesis to denoise the data,which according to the principle of point cloud registration.The hybrid registration method is used to select the standard target ball as the vital feature point to realise the record and splicing of the two-site cloud data.