Research On Progressive Classification Method Of Pole-Like Objects In Mobile Laser Scanning Point Cloud Data

Pole-like objects on both roadsides are important urban infrastructure.The rapid update of its information provides efficient data support for smart cities,intelligent transportation,and smart gardens.Mobile laser scanning system can quickly scan the objects from the side to obtain the point cloud data.The point cloud data provides a new data source for pole-like objects extraction and classification.Therefore,it is of great significance to explore an accurate extraction and classification method for pole-like objects based on mobile laser scanning point cloud data.The main problems of pole-like objects extraction and classification in mobile laser point cloud data are as follows.The terrain fluctuations are large,the noise interference is much;the pole-like objects are close to each other,and they cross each other;the shapes of the same kind pole-like objects are inconsistent.The existing methods are not very applicable to different data having the above problems.Based on the summary of the existing research methods,this paper proposed a progressive classification method for pole-like objects in mobile laser point cloud data.The main research work is as follows:1.The laser point cloud scanning profile and 3D shape of different objects were analyzed in depth,and the pole columnar part extraction method based on arc feature was researched.The algorithm firstly designed the non-pole columnar part point cloud denoising based on the scan line index.Then,the constrained RANSAC algorithm was used to extract the arc-like point set of the pole columnar part,and the 3D feature statistical law was further introduced to improve the anti-noise of the point set searching.2.There are different positional relationships between pole-like objects,such as separation,adjacent and intersecting.Based on the positional relationships and grid index,a ring-shaped neighborhood azimuth coverage evaluation factor was proposed.This factor focuses on the distribution differences in different positions of different pole-like objects point clouds.Different extraction conditions were automatically designed in the case where the pole-like objects are close to each other and occluded from each other.The overall target point cloud of trees and man-made pole-like objects can be accurately extracted separately.3.According to the unidirectional extension commonality and 3D morphological difference of the upper point cloud of different man-made pole-like objects,the differential quantitative classification method under the constraint of the principle direction was researched.A unified local coordinate reference with the principle direction as the constraint was established.Under this reference,the differences between different man-made pole-like objects are carefully quantified by eigenvalue analysis and projection analysis.Accurate classification was achieved for different types of pole-like objects and same types of pole-like objects with different dimensions.In this paper,the individualized point cloud data of different kinds of pole-like objects are finally extracted.Through the analysis and quantitative evaluation of the experimental data of two sections and four typical sets,it is verified that the proposed method has better extraction and classification effects on different pole-like objects in different road environments.