Laplacian Fusion Approach Of Lidar And Photogrammetric Point Cloud

In the construction of digital twins and smart cities,oblique photogrammetry and laser scanning are the two main ways of geographic information collection.Single source of data is difficult to meet the needs of increasingly complex three-dimensional urban spatial perception and application.Data fusion is the main method to enhance the ability of scene expression.Through multi-view stereo matching and other methods,data from different sources can be converted into the most basic point primitives,and modal differences can be eliminated,so that the effective fusion of multi-source point clouds become the basic way for improve the quality of urban scene expression and realize the atomized twin of the real world.There are significant differences between the above two types of point clouds in terms of spatial resolution,coverage,and noise,which leads to the difficulty in expressing the same scene and feature spatial location,description scale,and semantic information.Thus extracting traditional structures and spatial benchmarks.Fusion tasks such as structure extraction and spatial datum consistency integration have presented new challenges.This article focuses on the fusion of laser point cloud and image-based point cloud.In view of the challenges brought by the characteristics and significant differences of different types of point cloud data,explore a consistent fusion method for large-scale urban environment scenarios,adapting to complex scene structures and feature types.Innovatively propose a multi-source point cloud fusion full-process processing solution that takes into account multiscale geometric features,and recognizes stable and accurate geometric structure features from multi-scale.From the three aspects of the alignment of the spatial datum,the difference processing of the overlapping area and the wide gap processing of the connecting area,the accurate extraction and explicit expression methods of the main geometric structure and the local neighborhood structure are applied.Improve the consistency of the data after fusion,with higher coverage,higher accuracy,and richer details,which significantly improves the quality of expression of urban scenes.This will further support smart city applications such as 3D modeling,urban planning,and change monitoring,and provide new methods and ideas for more modal data fusion solutions.The specific research content is as follows:(1)Incremental registration method for multi-source point cloud considering main geometry structureThis paper proposes an incremental point cloud registration strategy,which applies the main geometric structure of buildings and artificial structures in the scene,and realizes the registration of multi-source point clouds by the interrelated two-dimensional transformation.It overcomes the difference of multi-source point cloud and noise image,reduces the complexity and automatically restores the scale.This method realizes cross-source point cloud registration for multiple platforms in the air and on the ground.Through qualitative and quantitative analysis,the accuracy of automatic recovery scale reached 98.5%,and the registration accuracy could reach centimeter level under the influence of noise interference and significant difference,showing good applicability and accuracy.(2)Point cloud cleaning method of overlapping region based on noise recognition and voxel structure extractionA point cloud cleaning method for overlapping areas is proposed to deal with the contradictions such as layering,cracks,and surface intersections between point clouds caused by the differences in the description of the same scene object from multiple source point clouds.Through adaptive spatial voxel division,the complex scene is divided into simple spatial units that can be processed independently.Based on accurate spatial mapping and accurate extraction of voxel structure,a multi-dimensional geometric feature descriptor is constructed to realize noise recognition and redundant cleaning of overlapping areas,and improve the integrity and accuracy of the fusion point cloud.Through qualitative and quantitative experiments on the two sets of data,it can be seen that the cleaned fusion point cloud set has higher integrity and accuracy,the number of effective point clouds increased by 23.77%,and the error decreased by 15.00%,which well proves the effectiveness of the method.(3)Gradient domain Laplacian repair method of multi-source point cloud gapThe hollow area of the point cloud can be filled with multi-source data,but the differences between the multi-source point clouds lead to the problem of wide gaps such as cracks and dislocations.This paper proposes a Laplacian repair method in the gradient domain.Its core is the innovative gradient domain Laplacian coordinate point cloud description,which explicitly expresses the local structural details of the point cloud,so as to ensure that the internal details of the cavity are not lost during the wide gap repair process.Through accurate adjustment of boundary anchor points and coordinate conversion between different domains,the cracks are repaired,and the details inside the cavity are not lost.Through abundant qualitative and quantitative experiments,it is proved that the method proposed in this paper can effectively repair the wide gap phenomenon of various types of point clouds between sources,and can maintain more than 75% accurate repair for the most complex irregular geometric layout holes,improving the integrity of data and ensuring the rich details of filling data.