| With the application of 3D point cloud data in many domains widely,the processing techniques of 3D point cloud data has become a hot spot of research at home and abroad.As a key technique of data processing,feature extraction from 3D point cloud data is the basis of the follow-up processing,such as region segmentation,surface reconstruction and so on,and has a very important effect on actual applications.Feature extraction from 3D point cloud data refers to extracting various kinds of geometric discontinuity points from 3D point cloud data,such as discontinuity of zero order(~0),discontinuity of first order(~1),and discontinuity of second order(~2).These geometric discontinuity points describe important sharp feature of object surfaces.Therefore,it is very important for improving the application level of 3D point cloud data to research feature extraction accurately.This paper proposes a novel method based on local outlier factor algorithm(LOF)to extracting zero order discontinuity points and first order discontinuity points from regular 3D point cloud data.Firstly,the proposed method estimates the geometric properties of surfaces and researches behaviors of chord lengths,angle variations of the unit tangent vector,and curvatures at the geometric features.The LOF algorithm is used to extract the extremum of chord lengths and curvatures,then zero order discontinuity points and first order discontinuity points can be identified.In the process of using LOF algorithm,the distribution of local outlier factor values is analyzed by a boxplot and then the threshold value can be set.The proposed method is reasonable,efficient and feasible.In order to test the performance of our method for extracting geometric discontinuity points from 3D point cloud data,we make the feature extraction experiments on regular 3D line point clouds and 3D point clouds.For 3D line point clouds,this proposed method is used to extract zero order discontinuity points and first order discontinuity points without noise and with noise respectively.In view of 3D point clouds,we extract four groups of line point clouds:the horizontal line point clouds,the vertical line point clouds,the sloping-up line point clouds and sloping-down line point clouds along four direction from the structured 3D point clouds,then using our method to extract zero order discontinuity points and first order discontinuity points from line point clouds,the last four groups of extracted discontinuity points are merged together and these merged discontinuity points are the geometric features of the 3D point cloud data.Experimental results show that the proposed feature extraction method is valid and reliable. |
