Research On Vegetation Removal From Point Cloud Data Of Bedrock Slope Based On Dimensional Characteristics

As a new mapping technology,Ground Lidar is widely used in slope deformation monitoring,reverse engineering,power line cruise,three-dimensional modeling and other work.In order to obtain more accurate slope coordinate information in deformation monitoring,it is necessary to remove vegetation points in natural scenes as far as possible.However,in previous studies,the methods used by scholars to classify point clouds have some problems,such as low universality and randomness in accuracy.In order to improve the classification effect,according to the theory of machine learning and statistical analysis,this paper uses the point cloud classification algorithm based on multi-scale dimension characteristics(MSDF)and the outlier removal algorithm based on distance and statistical distribution(OR-DSD)to remove the vegetation point clouds in the point clouds of bedrock slope obtained by three-dimensional laser scanner,and from the selection of classification scales,outlier detection and removal,error deletion point recovery.The algorithm is optimized and supplemented in complex aspects to improve the point cloud accuracy of rock slope.An example of Sihetang village dangerous rock slope in Miyun District of Beijing is given to verify the algorithm.The main achievements are as follows:(1)According to the characteristics of point clouds on vegetation and bedrock slopes,an automatic classification method of point clouds under complex terrain conditions is determined,which is a point cloud classification algorithm based on multi-scale dimension features.By using obfuscation matrix to evaluate the accuracy of classification results under different conditions,the rules for selecting parameters of MSDF based on "partition block classification,small scene first" are proposed,which reduces the influence of selecting parameters only by experience and shortens the time required for classification.(2)According to the spatial characteristics of point clouds on slopes classified by MSDF,outlier detection algorithm is used for secondary removal.A method to characterize the dispersion of point clouds by standard deviation of point distance is proposed,and the neighborhood size used in outlier detection is determined.Therefore,outlier detection and removal process of "rough detection first,then fine detection" is proposed.An angle-based edge point recognition algorithm is introduced to recognize the edge of the main point cloud after removing outliers.The region growth is centered on the obtained edge points,and the point cloud around the boundary which was deleted by mistake in outlier detection stage is restored.(3)Vegetation removal experiments on Sihetang village slope point clouds obtained by three-dimensional laser scanner were carried out.The results show that the average accuracy of slope point clouds obtained by optimal scale classification is 98.52%,and the average accuracy of slope point clouds obtained by outlier removal and error deletion recovery is 98.87%.This shows that in the process of removing point cloud vegetation on complex bedrock slopes,the accuracy of the above method is higher than that obtained by MSDF alone.