| As a typical feature in the urban street system,street trees have important ecological functions in improving urban environment,containing water and purifying air.Therefore,street tree resource survey is an important part of urban ecological resource survey.Traditional street tree information acquisition mainly adopted such means as manual survey,total station measurement,RTK measurement,etc.,which were inefficient and heavy workload,and could not efficiently met the needs of planning and management departments to monitor urban ecological information.Mobile laser scanning technology,as an advanced measurement means,can quickly collect the point cloud information of street features.The point cloud data obtained by mobile laser scanning system can be used to extract street trees and attribute information with high efficiency,high accuracy and strong timeliness.The spatial distribution of street tree point clouds is complex without topology,and it is challenging to extract street tree and attribute information from street point cloud data.In this paper,three research works are conducted around this problem as follows.(1)A random forest model-based point cloud detector for street trees is trained.The method separates the street tree point clouds from the complex street point clouds by establishing the k-d tree index,querying the spherical neighborhood nearest neighbor points,constructing the multidimensional feature vector,creating the integrated learning model,and classifying the detector.The experimental part uses mobile laser scanning point cloud data for experiments and evaluates the classification results metrically.The results show that the method can effectively detect the street tree point cloud with 97.82%Precision,97.97%Recall,and 97.89%F1 score on the test set.The method uses a spherical neighborhood that flexibly adapts to various scenarios in the step of constructing multidimensional feature vectors for local point clouds,which can be applied to point cloud data in complex street environments.(2)A connected canopy segmentation and optimization method based on trunk marker point localization is proposed.First,before the initial segmentation,the street tree point cloud data set is pre-processed with down-sampling and coordinate correction;second,a density clustering-based method is used to process the street tree point clouds to segment the non-adhesive street tree point clouds from each other;then,for the case that some street tree point clouds have connected canopy point clouds,a trunk marker point positioning method is used to segment all single street trees;finally,a combination of k-nearest neighbor and density Finally,a combination of k-nearest neighbor and density clustering is used to optimize the attribution of the treetop point clouds near the segmentation surface.The single tree segmentation method fully considers the computational cost,the original point cloud data skew and the canopy adhesion of the street trees,and is highly applicable.(3)The method of extracting attribute information of street trees is improved.To address the problem of deviations in the attribute information measurement of some street trees caused by the flatness of the ground,the normal vector of the ground point cloud near the rootstock of the street trees is extracted,and the single wood point cloud data is corrected based on this normal vector.In order to avoid the influence of street tree drooping branches on the measurement accuracy,a k-nearest neighbor learning method was used to extract the trunk point cloud data from the bottom up,and the RANSAC algorithm was used to fit the model,and the extracted chest diameter parameters were compared with the manual measurement data,with the regression evaluation index~2=0.9441 and=0.026m.The histograms of the distribution of the measurements of the two attributes,crown width and tree height,were also consistent with the characteristics of the tree species at the experimental site. |
PDF Full Text Request