Researching The Filtering And Clasification Algorithms Of Airborne LiDAR Point Cloud

With the rapid development of spatial information technology, the requirementabout the ability to obtain geospatial data continue to increase, traditional spatialinformation acquisition technology can’t meet the actual needs, the means ofMulti-platform、Multi-sensor to get information draws more and more people’sattention, and the satellite borne、airborne platform has become the important way toget High-tech data. As an active ground observation system, Airborne Light Detectionand Ranging system which are integrated Global Navigation satellite system、inertialnavigation system、laser ranging system and imaging device.,and because its highprecision、high speed、the transmission ability,as well as the directly obtainthree-dimensional coordinated of the Earth surface ability, this technology has beenwidely used to obtain large area digital terrestrial model and elevation model.Although the hardware of Airborne Light Detection and Ranging has beencontinually improving, the problem of system integration has got a better solution, butdata processing technology is lagging behind. After a lot of exotical and domesticalscholars’s researches, they put forward many point cloud filtering and point cloudclassification algorithms, but the automation degree of these algorithms is not high,because these algorithms involve too many parameters with complex processingprocedure. To overcome the deficiencies above, this paper proposes a denoising、filtering and classfication algorithms based on the density of cloud point.The maincontent and research work are summarized as follows:(1) This paper make a introduction of the method to calculate point cloud density,through the establishment of the convex hull of point cloud, and then using vectortriangle area to calculate the area of the convex hull, the ratio between the obtainedpoint cloud and convex hull area is point cloud density. By using point cloud density,we can get the distance between different points.(2)This paper researches the denoising algorithm of point cloud data, TheDistance to the Mean Elevation and Adaptive Moving-Box denoising algorithm. TheDistance to the Mean Elevation algorithm is firstly worked through calculating meanelevation of the area, then calculating the distance between each point and the meanelevation,finally arranging the result from small to large,.if it has noise, then it cangenerate a large negative or positive distance,excluding the large distance points inboth ends to the achievement of point cloud denoising; Self-adaptive Moving-Boxdenoiding algorithms is based on the relevance of spatial objects, on the basis of the known point cloud density,calculate the adaptive size of box to achieve cloud pointdenoising. Experiment shows that this denoising algorithm has a better denoising andstrong application effect.(3) On the basis of denoising,this paper design a terrain correlation coefficientpoint cloud filtering algorithms. In order to increase the relevance of scatter pointcloud and exclude the point belongs to building better, this algorithms interpolationTIN to approximate the terrain configuration after sorted by elevation.In the processof building TIN model, we can calculating the terrain correlation coefficient betweennew triangles which was generated by the unknown inserted point and topologyadjacent triangles、the distance between unknown point and interpolatin triangle aswell as the maximum terrain slope parameters, to judge whether the pending pointsare ground points. Proofed by relevant experiment and analysis, this filteringalgorithm has a good filtering effect and strong adaptability.(4) On the basis of filtering, this paper design a point cloud classificationalgorithms which is based on surface roughness. The non-ground points would clustedby distance after been filtered, after clustering, each point on behalf of a particularfeature, then calculate the minimum area for each class of objects、elevation amplitudeand surface roughness to determine the class categories, finally realizing theclassification of point cloud. Proofed by relevant experiment, this classificationalgorithm has a quite accurate result.