Landscape Tree Extraction Based On Marked Point Process With Spatial Feature Constraints From LiDAR Point Cloud Data

In the digital age,the information contained in the geographic information system is becoming more and more abundant,and it has gradually developed to three dimensions.Three-dimensional reconstruction is a very important task in the construction of digital cities.As the main spatial entity in the city,the accurate extraction about three-dimensional structure information of landscape trees can provide a more accurate data basis for the construction of ecological cities.At present,most of the researches on the extraction of urban Li DAR(Light Detection and Ranging)point cloud data are based on buildings,and there are relatively few studies on the extraction of landscape trees.Airborne Li DAR is a new type of remote sensing technology that can directly and quickly obtain high-precision three-dimensional information ground and surface objects.It is suitable for rapid and accurate measurement of urban three-dimensional spatial information.It has unique advantages in information collection for large-scale urban areas and is widely used in basic surveying and mapping,urban mapping and3 D modeling.Based on the geometric structure information of the targets of interest in the Li DAR point cloud data,this paper makes full use of spatial information and spatial relationships to construct a MPP(Marked Point Process)model to describe the spatial distribution of the targets,and proposes a new algorithm for landscape trees extraction from Li DAR point cloud data based on a MPP with spatial feature constraints.This algorithm separates non-ground points from Li DAR point cloud data,and constrains the MPP model through the canopy point cloud features.Firstly,the triangulated irregular network processing densification filtering algorithm is used to separate the ground points and the non-ground points from the Li DAR point cloud data.Then,the circle is used to describe the geometry of landscape trees in the ground projection area to define the marked point process model of spatial distribution of landscape trees,in which the random point process defines their locations and the circle marks associated to the points indicate their geometric structures.Elevation distribution model is built by combining the characteristics of elevation distribution of landscape trees and non-landscape trees.Accordingly,an elevation constraint model is constructed with respect to the spatial density characteristics.By Bayesian theory,the posterior probability modeling landscape trees extraction is defined by integrating above circle marked point process model,elevation distribution model,elevation constrained model and correlative parameters distributions.The RJMCMC(Reversible Jump Markov Chain Monte Carlo)algorithm is designed to simulate landscape trees extraction model.Finally,the optimal landscape trees are obtained under the MAP(Maximum A Posteriori)criterion.Experimental results show that the overall accuracy of the algorithm is relatively high;the extraction of landscape trees can also achieve higher accuracy for complex scenes that are difficult to identify.According to the proposed algorithm in this paper,a number of typical urban areas are selected from the airborne Li DAR point cloud data set as the research area,and experiments are carried out respectively.The results show that the overall accuracy of the proposed algorithm for extracting landscape trees is higher,and the overall extraction rate and accuracy rate have reached more than 90%.The extraction results of landscape trees in complex scenes that are difficult to identify can also achieve higher accuracy.The paper has 19 diagrams,9 tables and 61 references.