Classification And Extraction Of Mobile LiDAR Point Clouds In Urban Areas

Laser Radar technology(LiDAR,Light Detection and Ranging)is a new solution for 3D geospatial data acquisition by scanning and detecting the objects' detailed 3D spatial information from far away.It has an advantage of a direct,fast and accurate measuring way,as is widely used in environmental monitoring,3D modeling,urban planning and other fields.Mobile LiDAR system takes the vehicle as carrying platform,combining laser scanner,GPS,IMU and CCD camera.As the vehicle moving,it can capture the surface spatial information of all scanned and detected target objects,then high-density,high-accuracy and rich scene of point cloud data is obtained.However,as blindness of scanning,the phenomenon of noise points and empty data exists,which increases the burden of data post-processing.This paper mainly studies point cloud data filtering and feature extraction around the theme of classification,the main research content is as follows:1.The components Mobile LiDAR system,point cloud data characteristic and data organizing mode are introduced.The spatial distribution characters of several typical objects are analyzed,then classification strategy and data processing flow are put forward by considering algorithm principle and key problems of existing filtering and feature extraction technology.2.Removing noise points in original point cloud data,local elevation difference of K neighboring points is used to extract preliminary ground points.Aimed at the confusing case of K neighboring points,local planar feature is utilized as constraint.Through analyzing error precision,it is proved that the method has some validity.3.A method of utilizing elevation slice assist shape feature to recognize the vehicle is put forward in paper,set sliced three-dimensional target projected to two-dimensional gray image according to elevation information and do image binarization.Combined with shape constraint conditions,an improved expansion algorithm is used to connect and detect the target areas.4.A new method of separating tree points is proposed based on dispersion degree of local spatial distribution,utilizing the difference in horizontal projection area size of different objects to separate tree points.Utilizing the containment relationships between trunk points and canopy points to extract unclassified trunk points.5.Utilizing depth image to identify buildings based on the obvious difference in elevation of buildings and other objects.In connection with the leakage points of low parts,the PCA technology is used to complement residual building's facade points based on the dimension feature of planar data points.6.The streetlights have typical character of certain height,regular shape,the maximum elevation difference and the density in the same virtual grid are used to extract lampposts.Aimed at the lack of lamp holders imformation,to make circle search of lampposts and the lamp holders are supplemented.