Classification Of Airborne LiDAR Point Cloud Based On Random Forests

In recent years,airborne Li DAR has been widely used in many fields such as terrain mapping,3D modeling,and urban planning because of its high-precision,high-efficiency data acquisition.However,the data acquired by airborne Li DAR is disorderly 3D point cloud data,and can not directly obtain the three-dimensional information of the ground and objects.Therefore,the research on the efficient and high-precision point cloud classification method has become the current research hotspot.According to the basic unit of feature extraction,the existing point cloud classification methods can be classified into point-based classification methods and segment-based classification methods.Point-based classification methods classifies the point cloud data by analyzing the characteristics of a single point.Segment-based classification methods first uses the point cloud segmentation algorithm to segment the point cloud data,and then classifies the point cloud data based on those segments.Compared with point-based classification methods,segment-based classification methods can obtain more abundant information and improve the classification effect.However,the existing segment-based classification method not only has an problem of unreasonable point cloud segmentation,but also ignores the influence of unreasonable features on the classification effect.This thesis studies the segmentat-based classification method and proposes a new classification method.The classification method is divided into three stages,namely,stepwise point cloud segmentation,point cloud classification based on random forest,and optimization of classification results.In the stage of stepwise point cloud segmentation,in allusion to the problem of over-segmentation and under-segmentation of the region growing algorithm,the random sample consensus(RANSAC)algorithm and the region growing algorithm are combined,and a patch-based region growing algorithm is propose,and at the same time,in allusion to the weakness of a single point cloud segmentation algorithm,we combines the region growing algorithm and the euclidean clustering algorithm to segment the point cloud data into three kinds of segments.In the stage of point cloud classification based on random forests,considering the influence of unreasonable features on the classification effect,this thesis uses random forests to select features,which improves the robustness and stability of the classification method.In addition,in the optimization stage of classification results,this thesis proposes an optimization method that combines semantic rules and k-nearest neighbors(KNN)algorithm.In the optimization process,this thesis first uses a conditional euclidean clustering algorithm to segment the point cloud data into objects.After that,this thesis use the semantic information of objects to set the corresponding semantic rules,and use these semantic rules to detect each object and find unreliable objects.Finally,this thesis use the KNN algorithms to classify point cloud data that is unreliable objects.This improves the effect of point cloud classification.This thesis tests the classification methods proposed in two test data sets.The experimental results show that the point cloud segmentation method proposed in this thesis can effectively segment the point cloud data,and the point cloud classification method can achieve better classification results.