| As the basic way of power transmission,transmission lines are important energy infrastructures,which are related to the national economy and national energy security.As an active remote sensing technology,airborne LiDAR can directly and quickly obtain high-precision,dense 3D points without being restricted by light and terrain.It is widely used in power inspection to ensure the safe power supply of transmission lines and provide basic three-dimensional data for building smart grids.Due to the large number of point cloud and their discrete structures,simulation analysis and model visualization cannot be performed directly.They need to be converted into a high-precision and refined three-dimensional model to serve the smart grid platform.A series of data processing is required from the original point cloud to the real 3D model.Among them,the pylon positioning algorithm has poor universality,the best classification radius of power elements is unclear,the precise pylon point cloud extraction algorithm is insufficiently studied,and the pylon reconstruction algorithm is inefficient.Based on the existing research,this paper improves the pylon positioning algorithm,clarifies the optimal classification radius of power elements,improves the purity of the pylon point cloud and the accuracy and efficiency of the pylon reconstruction algorithm,from the point cloud segmentation of transmission lines to the generation of 3D models of power elements.And the measured data is used to verify the reliability of the proposed algorithm.The main contents and conclusions are as follows:(1)Improving existing automatic power pylon positioning algorithm.Firstly,based on the PCA algorithm,the horizontal orientation of point cloud of transmission lines is redirected to improve the utilization of the two-dimensional grid.Then the existing grid growth rule is improved,make full use of the macro characteristics of the area where the power pylons and power lines are located,and use the maximum height of the grid as the growth parameter to determine the candidate power pylon and power line areas.After that,a new grid class cluster discrimination threshold is introduced,and the power line region is determined by the class cluster horizontal contour coefficient and the maximum projection length.Finally,based on the characteristics of the existing vertical continuous distribution coefficient,a grid elevation distribution coefficient and a grid convex hull coefficient were added to realize the automatic positioning of the pylon coordinates.Point cloud of three high-voltage transmission lines with different pylon heights and terrains were selected as experimental data.The experimental results show that the algorithm has an average precision rate of 100%,an average recall rate of 100%,an average F1 value of 100%,and average efficiency of4.1s/km,which has good universality.(2)Research on point cloud classification algorithm based on machine learning.Voxel features are used to construct 16 features and combine them into 4 feature sets.Two types of neighborhoods(spherical and cylindrical neighborhoods)and seven neighborhood radius(1m ～ 4m)are used to calculate each type of feature set for training the RF classifier.It adopts 4 test scenarios and conducts 112 classification experiments.Based on the experimental results,the feature set and feature sensitivity are compared and analyzed systematically.The analysis results show that:(1)The classification performance of the feature set(SE,OV,EE,AS,PL,LI,SP,RO,VE,DE,LVV,VH,VHC,and VV)is the best,the optimal neighborhood radius of the comprehensive classification is 2m,the average precision rate of each classification is94.8%,the average recall rate is 82.9%,the F value is 87.1%.And the best neighborhood radius of power element classification is 3.5m,the average precision of power lines and pylons is 98.9%,the average recall rate is 96.3%,and the F value is97.6%.(2)Feature LVV,VHC,RO,OV,DE,VH,SE,VE and LI are not sensitive to different radius.Among them,LVV,VHC,and RO are strong classification features,and the classification contribution is large.OV,DE,VH,SE,VE,and LI are weak classification features,and the classification contribution is small.SP,PL,and AS are sensitive features,and they are more suitable for classification of ground,vegetation and buildings.(3)Based on the unique structural characteristics of the pylon,a refined pylon extraction algorithm based on structural constraints was proposed.Firstly,the “groove”at the bottom of the pylon is eliminated by the downward spatial grid 17 neighborhood growth algorithm.Then,based on the geometric features of the redirected plyon body,part of the plyon body point cloud was obtained.Finally,for the upper point cloud of part of the plyon body,the upward two-dimensional grid 5 neighborhood growth algorithm and the method based on the structure constraints of the frustum pyramid are used to remove the attachment component points.For the lower point cloud of part of the plyon body,the method based on the constraint of the frustum pyramid structure and the inverted triangular pyramid structure was used to eliminate the mistake points outside and inside the pylon.Six different pylon types,attachment components,and mistake points distribution of pylon points were used for experiments.The experimental results show that the average precision of the refined extraction of the pylon is 98.5%,the average recall rate is 96.7%,and the average F1 value is 97.6%,the average time consuming is 0.9s,and the parameter stability is good.(4)Based on the generalized structural characteristics of the pylon,all the pylon structures were subdivided into inverted triangular pyramid structure,columnar structure,quadrangular frustum pyramid structure and complex structure for the first time,and an automatic reconstruction algorithm of the pylon based on the template structure was proposed.Firstly,based on the existing point density feature,the filling rate feature is introduced to improve the accuracy of division position recognition and pylon type recognition.Then,the columnar structure,inverted triangular pyramid structure and q quadrangular frustum pyramid structure were reconstructed based on the generalized concrete template structure.For the quadrangular mesa structure,the internal structure reconstruction is added on the basis of the existing reconstruction.Based on the generalized abstract template structure,the topological relationship of the complex structure model is determined,and the 3D point cloud and 2D image data processing method are used to reconstruct the complex structure.The mathematical model is used to optimize the corner coordinates to solve the corner "swell" caused by image processing.Twelve pylon-type point cloud are used to verify the proposed algorithm.The experimental results show that the average accuracy of the 12 pylon models is 0.140 m and the average time consuming is 0.8s.The parameters used in the algorithm are stable and have strong ability to resist the lack of pylon point cloud.(5)For the first time,a reconstruction method of insulator strings based on the power pylon and line model was proposed.By analyzing the characteristics of the insulator string connection structure in the transmission line,a endpoints of power pylon and line matching model is established,and the insulator string is reconstructed based on the candidate endpoints in the pylon model and endpoints in the power line model. |
