Study On Building Extraction Based On Airborne LiDAR Point Cloud Texture Feature

Automatic extraction and 3D reconstruction of buildings has been a hot topic in the field of photogrammetry and remote sensing.Airborne lidar data has become one of the most popular data for building extraction in recent years.Although Airborne LIDAR point cloud data can provide 3D information of ground objects,the discreteness of point cloud will lead to the loss of some correlation features;It is difficult to accurately extract building point cloud from vegetation regions with similar elevation;Considering the different texture features of each object surface,the boundary between textures can well correspond to the boundary of the target area in the image.Therefore,in this paper,the point cloud texture features are used to extract the building point cloud,and the texture is used to distinguish the different features in the point cloud data which are similar to the building elevation,so as to extract more accurate building point cloud.In this paper,the international standard building extraction experimental data provided by the third working group of International Society for Photogrammetry and remote sensing(ISPRS)is used.The main research work of this paper is as follows(1)Texture features are generated based on point cloud elevation map and intensity map.In this paper,four methods(GLCM,LBP,Gabor filter and WT)are used to generate texture features based on point cloud elevation map and intensity map respectively.The experimental results show that the performance of texture features generated by different methods is different in building and non-building areas,and the texture generated by point cloud elevation map is more obvious than that generated by intensity map in building and non building areas.(2)Building point cloud extraction based on point cloud texture features.Five feature sets(JH,JH + GLCM,JH + LBP,JH + Gabor,JH + WT)are composed of the geometric features of the point cloud combined with four texture features respectively.The feature combination of 40% ? 100% feature proportion is selected to extract the building point cloud.The experimental results show that: 1)the maximum building extraction accuracy of JH + GLCM,JH + LBP,JH + Gabor and JH + wt feature sets is higher than that of building extraction only using point cloud geometric features(JH).Adding texture features can reduce the error rate and missing rate of building extraction only using geometric features,and improve the edge information recognition effect of some buildings;2)Compared with Gabor and WT texture features,GLCM and LBP texture features show better building extraction results in the complex area of building top surface.The accuracy of building extraction by combining GLCM texture and LBP texture feature sets is generally higher than Gabor texture and WT texture sets;3)The four texture features based on elevation map are more conducive to building point cloud extraction than the four texture features based on intensity map,and have lower missing extraction rate and error extraction rate.(3)The height difference and area constrained region growth algorithm is used to post process the initial extracted building point cloud.This method aims at the missing extraction and wrong extraction of building point cloud in the initial extraction results,and refines the building extraction results.The experimental results show that the average integrity rate of building extraction is 90.1%,the average accuracy rate is 97.0%,and the extraction quality is 89.3%;This method can effectively eliminate the false extraction fragments in the initial extraction results,and the post-processing building point cloud results can be used for subsequent applications with a little manual filling in the small area of missing extraction building areas.