The Study Of Substation Equipment Extraction And Recognition Algorithm Based On 3D Point Cloud

3D reconstruction of substation scene is the basis for the construction of smart substation.With the development of virtual reality technology,substation reconstruction based on 3D point cloud has gradually become a research hotspot.However,current methods of substation modeling based on point cloud mainly use professional modeling software,such as Geomagic Studio,to generate substation 3D model interactively,which exists problems of low efficiency and precision.At the same time,the point cloud data obtained by laser scanner may have noise,occlusion and uneven density,which also increases the difficulty of substation modeling.Therefore,it is of great significance to study automatic substation reconstruction technology based on 3D point cloud.In this thesis,two important technologies of substation automatic modeling based on point cloud are studied in depth,including how to extract all independent equipment point clouds from substation point cloud data and how to recognize the extracted equipment point clouds.The main contents are as follows:(1)Preprocessing of substation point cloud data.Firstly,statistical filtering method is used to denoise the substation point cloud data.Secondly,octree coding method is used to simplify the substation point cloud data.Finally,the random sampling consistency algorithm is used to fit plane to achieve the removal of ground points.(2)Substation equipment extraction based on 3D point cloud.This thesis proposes an automatic extraction algorithm of substation equipment point clouds based on voxel growth.First of all,the point clouds of bus pipes adhered to substation equipment are removed by random sampling consistency algorithm and the remaining substation point cloud data is voxelized to generate substation point cloud voxel space.Secondly,the voxels lie in the bottom of the voxel space and meet certain conditions are selected as seed voxels to determine a seed voxels stack.Then,seed voxels are randomly selected from the seed voxel stack and start growing to realize coarse extraction of equipment point clouds.Finally,the missing points in the stage of coarse extraction are reunited,and the point clouds of wires adhered to the equipment are removed by principal component analysis to complete fine extraction of equipment point clouds.The algorithm proposed in this thesis is tested on a substation point cloud data provided by cooperative company.Experimental results show that the algorithm can successfully extract all the equipment point clouds from the substation point cloud data.Recall rate reaches 99.76%and accurate rate reaches 99.09%.Compared with another equipment point clouds extraction algorithm,the effectiveness of the proposed algorithm is further verified.(3)Substation equipment recognition based on 3D point cloud.Firstly,the symmetry and distribution density of equipment point cloud are used to establish a local coordinate system.Then,a new feature descriptor is defined for the description and recognition of substation equipment point cloud based on the shape of equipment point cloud and the differences in equipment views.Next,template matching is performed between the feature descriptor of the equipment point cloud to be identified and the Finally,129 equipment point clouds to be identified are used to test the algorithm proposed in this thesis and another two equipment point cloud recognition algorithms.Experiment results show that the average time of the proposed algorithm in this thesis to recognize one equipment point cloud is 3.2s,and the recognition accuracy reaches 90.7%,which demonstrates the effectiveness and efficiency of the proposed algorithm in the thesis.In addition,it can still ensure high recognition accuracy when the density of equipment point cloud is heterogeneous.