Research On Filtering And Classification Algorithms Of 3D Point Cloud Data

In recent years,the hardware of LiDAR has been continuously developed and improved,but the research on the algorithm of point cloud data processing has lagged behind.At present,scholars at home and abroad have carried out a lot of research and proposed many algorithms for LiDAR point cloud filtering and classification,but these algorithms are difficult to apply to various terrain because of their high requirements for terrain.For example,in mountainous terrain,the difference between the slope of the ground object and the slope of the terrain is small.It is easy to cause excessive filtering or the loss of the real terrain only a single slope threshold.In view of the above problems,this paper analyses the denoising methods of LiDAR point cloud data,and studies in detail the new methods of LiDAR point cloud data filtering and classification.The main work and research focus of this paper are summarized as follows:In order to overcome the problems of low automation and unsuitability of traditional filtering algorithms for various terrains,an adaptive point cloud filtering algorithm based on gradient block is proposed in this paper.The main contents of the algorithm are as follows: Firstly,the denoising method of LiDAR point cloud data based on R neighborhood is studied,and the outliers in point cloud are filtered out.Secondly,the filtering process of LiDAR point cloud is studied.Considering the defect of traditional slope filtering algorithm in the threshold setting,this paper improves on two aspects.First,in order to reduce the influence of terrain on the accuracy of the algorithm,the data block is added.In addition,the LiDAR point cloud is divided along the directions of x-axis and x-axis.Second,in order to obtain the optimal threshold for filtering between the ground point and the non-ground point of each LiDAR point cloud,an adaptive threshold is added.The proposed scheme improves the traditional maximum inter-class variance method,so that it can be applied to the filtering algorithm of 3D point cloud data.Finally,ISPRS data set is used to test the improved algorithm.The experimental results show that the improved algorithm reduces the error from 17.48% to 9.24% comparing with the original gradient algorithm,and can be applied to different terrains with good robustness.Due to the low accuracy and high complexity of classification algorithm,a Point cloud classification algorithm of Mixed Kernel Function SVM is proposed.Firstly,the features of LiDAR point cloud data are extracted,and the feature vector is constructed.Then a mixed kernel function of Gauss and polynomial is designed,and a one-versus-rest(OVR)SVM classifier is constructed.Finally,LiDAR point cloud data is sent to the classifier for training and testing,and the classified classes are output.The experimental results show that the proposed algorithm can better classify different types of point cloud data,and the classification accuracy is over 95%,and it has good robustness.In conclusion,the methods proposed in this paper have achieved better filtering and classification results,which lays a good foundation for the subsequent 3D reconstruction.