Technical Research On Mending Three Dimensional Point Cloud Holes With Images

When the three-dimensional laser scanner collects point cloud data,the point cloud data is incomplete due to the occlusion of itself or the surrounding environment,that is,there is a hole in the point cloud data.Incomplete point cloud data will affect the reconstruction of 3D models,so it is necessary to study the method of repairing point cloud holes.In this paper,photogrammetric techniques are introduced.By collecting solid image data around the location of point cloud holes,the image data is resolved into the original image point cloud data through the Motion Recovery Structure Algorithm(SFM)and Patch Matching(PMVS).The registration of the two kinds of point cloud data,the deletion of the original image point cloud data in the overlapping area and the optimal density selection of the image point cloud data complete the fusion of the two data 3D-2D,thereby completing the hole cloud data repair.The main research results of the paper are as follows:(1)When the image data is solved,the motion recovery structure algorithm(SFM)and the patch-based dense matching(PMVS)method are introduced in detail.The motion recovery structure algorithm(SFM)was used to extract and match the feature points in the image,Zhang Zhengyou camera calibrated the internal parameter matrix,and calculated the camera pose.Finally,the sparse three-dimensional point cloud data was generated using the triangulation method.Using dense patch-based matching(PMVS)to match,diffuse,and filter images to generate dense point clouds.(2)In order to improve the registration accuracy of the two kinds of point cloud data,this paper uses the Bursa algorithm in the seven-parameter algorithm to complete the fine registration of two types of point cloud data,focusing on the Bursa algorithm centered on gravity,and the method of selecting the initial value of the scale.Experimental results show that this method can effectively complete the accurate registration of the two kinds of data.(3)In order to reduce the point cloud data in the overlapping area,this paper uses k-proximity algorithm to find the corresponding point pairs in the two point clouds,and uses the distance between the corresponding point pairs to complete the screening work whether the point cloud is located in the overlapping area.The experimental results show that this method can effectively screen the point cloud data of the two point cloud data overlapping areas,and then delete the point cloud data in the original image point cloud data in the overlapping area.(4)Given that the accuracy of the image data is lower than that of the point cloud data,the optimum density of the image point cloud data is compared by the section method.First,the method of calculating the data density of point cloud is studied,and several methods of point cloud data reduction and point cloud data interpolation are introduced.Secondly,the minimum distance method is used to simplify the image point cloud data,and the nearest point interpolation method is used to interpolate the image point cloud data to obtain a variety of density image point cloud data.The cross section method compares the optimal density.The experimental results show that,considering the number of integrated point clouds,the best image point cloud data is 1.2 times that of the corresponding point cloud data.(5)In order to test the feasibility of the method of repairing point cloud data using image data,this paper uses the surface area evaluation and bias evaluation method respectively to combine the optimal image point cloud data with the point cloud data to be repaired and model the original point cloud data.Comparing the models established,the experimental results show that the area ratio obtained by the surface area method is less than 0.5%,and the maximum deviation obtained by the deviation method is 0.0036 m and the minimum is 0.0015 m.