Study On The Point Cloud Filtering Method Of Mudflat Topography Based On Hovercraft-Borne LiDAR

As mudflats contain abundant marine mineral and biological resources,they become hot areas for the comprehensive coastal development and utilization projects.The scientific exploitation of mudflats brings huge social and economic benefit.Obtaining the mudflat topography is the basis of the development and utilization of mudflats.The hovercraft-borne Li DAR(light detection and ranging)system can obtain accurate point cloud of mudflat efficiently,and the mudflat topography can be obtained after filtering.However,there are a lot of abnormal points under the ground and low object points close to the ground in the original point cloud of mudflat,which makes it difficult to filter the point cloud.Traditional point cloud filtering methods usually get wrong filtering results due to the outliers below the ground surface.Therefore,these outliers need to be removed manually before filtering,which reduces the filtering efficiency.In addition,traditional point cloud filtering methods do not make full use of the normal vector,intensity or other information of the point cloud,so it’s difficult for them to recognize ground points and low object points correctly at the same time.In order to solve these problems,this paper proposes an automatic point cloud filtering method,which considers the abnormal points below the ground surface and makes full use of the elevation,normal vector and intensity information of point cloud.The main work and contributions of this paper are as follows:1.The system composition and mechanism of hovercraft-borne Li DAR are introducedThe system composition of hovercraft-borne Li DAR is introduced;the operation process and data processing method of mudflat topographic survey based on hovercraftborne Li DAR are introduced,and the evaluation methods of filtering accuracy and point cloud accuracy are given;the error sources of point cloud coordinates are analyzed,and the method of estimating the valid scanning range of the system is given.2.The point cloud segmentation by combining normal vector and intensity is proposed.Firstly,the estimation methods of neighborhood fitting plane,normal vector and fitting residual of point cloud,as well as the preprocessing methods of intensity value of point cloud are given.Secondly,according to the three components of normal vector and intensity,the four-dimensional attribute vector of the point is constructed,and the calculation method of the attribute vector distance between two points is given to measure the similarity between two points.Finally,according to the distance from the candidate point to the neighborhood plane of the seed point and the attribute vector distance between the two points,the point cloud is segmented by the region growing algorithm.On this basis,the normal vector of point cloud is modified according to the segmentation results.This method realizes the complementation of normal vector and intensity information,solves the problem that the traditional point cloud segmentation method based on the difference of normal vector is easy to produce over segmentation,and improves the accuracy of point cloud segmentation.The experimental results verified the effectiveness and superiority of this method.3.The surface fitting method considering abnormal points below the surface is proposed.First,the cloth simulation algorithm used for surface fitting is introduced;then,based on the characteristic that mudflats are flat,the method of the second-time surface fitting on the basis of the first fitting is given.This method can automatically and accurately fit the mudflats surface with the consideration of the abnormal points below the surface.It solves the problem that the traditional surface fitting method is affected by the abnormal points below the surface and fits the wrong concave surface.It also improves the filtering efficiency because it does not need to delete the abnormal points below the surface manually.The experimental results show the effectiveness and superiority of this method.4.The comprehensive point cloud filtering method considering many kinds of information is proposed.Firstly,the point cloud is segmented by combining the normal vector and intensity information to obtain objects and the ground surface is fitted;secondly,the possible ground point is judged according to the distance from the point to the fitting surface and the angle between the normal vectors of the point and the fitting surface;thirdly,the ground object is judged according to the proportion of the possible ground points in the object;finally,all points in the ground object are determined as ground points.This method can filter out the low object points close to the ground surface as well as retaining the ground points because it fully considers the height,normal and intensity information of points.It solves the problem that the traditional filtering method which only considers the distance from the point to the fitting surface cannot retain ground points and filter out low object points at the same time and improves the filtering accuracy.The experimental results show the effectiveness and superiority of this method,including the advantages of the point cloud segmentation combining normal vector and intensity information,the modification of normal vector,and the constraint of normal vector angle.In the experiment,the traditional method achieves 3.45%,2.11% and 3.31% of type I error,type II error and total error,while the comprehensive filtering method considering many kinds of information in this paper reduces the corresponding errors to 0.53%,0.09% and 0.35%.5.Field experiment and data processingA topographic survey experiment was carried out on the mudflat of Liuhe River by a hovercraft-borne Li DAR system.Point cloud segmentation,surface fitting and comprehensive filtering are carried out for the acquired point cloud data,and the point cloud filtering accuracy and elevation accuracy of the ground points are evaluated.Finally,the total filtering error of 0.28%,the relative elevation accuracy of 3.1cm,and the absolute elevation accuracy of 6.4cm are obtained,which further verifies the hovercraftborne Li DAR system and the proposed data processing method can efficiently obtain high-accuracy mudflat topography.