Building And Road Extraction From LiDAR Data Based On Contour Feature Analysis

In developing LIDAR system, three-dimensional data from earth surface or object can be obtained directly. The LiDAR data, which is shown as point-cloud, has two organizing forms:irregular and dispersed. The three-dimensional (3D) information is obtained by end user with professional methods of data processing. Different purposes require different ways of data processing.Extracting buildings and roads from LiDAR point cloud is a hot research topic at present. Generally, the Filtering-Segment (F-S) algorithm is a usual extracting algorithm. The F-S algorithm has two steps. First, the LiDAR data is smoothed by a morphological-based filter or a surface estimation filter to generate the Digital Terrain Model (DTM). The difference between the DTM and the Digital Surface Model (DSM) derived from the LiDAR data may then be classified into ground data and non-ground data by an initial height threshold. Second, the buildings are segmented from the non-ground point clouds by geometric characteristics or fusing other information such as multispectral information, in which vegetation points may be removed for the better extraction of buildings. The roads are segmented from the ground point clouds by usually removing the vegetation canopies via the intensity information of LiDAR.At present, there are three weak points in the F-S method, i.e., application scope, accuracy rate and practicality. The F-S method is mostly suitable for the flat area, but not for the mountainous area. In addition, there still is some non-building or non-road data after building or road segmentation using the F-S method, which reduces segmentation accuracy rate. In the existing road-segment algorithms, intensity information is used as accessorial information. But not all Lidar systems can obtain the intensity information. Therefore, when without intensity information, the practicality is strained.This paper presents a new segment method which is called the Filtering Algorithm Based on Contour-Segment (Fc-S). First, the Fc-S algorithm is implemented mainly by filtering based on the characteristics of DSM contour, that is, on the basis of the characteristics of DSM contour, such as the closure feature or the distance between the start and end points of the contour line, the contour line which belongs to natural ground is extracted by threshold method automatically, then the initialized ground point cloud is acquired and interpolated to obtain initialized DTM. The refined DTM is generated by an iterative approximation method, i.e., Filtering Based on Contour (Fc). After the point cloud has been classified into the ground-and non-ground point sets, the height threshold, the edge information, the gradient and area threshold value are adopted to remove the vegetation points from the non-ground point cloud, and thus the building point cloud is extracted. The refined building points are obtained by such iterative approximation method. As the object area's spectrum information can be acquired by digital camera, the roads may be extracted from the spectrum information. The parking lot and other areas that have a spectrum similar to that of the road could be removed by the geometric characteristics of road.To improve the efficiency of building extraction by the Fc-S algorithm, the Contour Shape Analysis (CSA) is used. Based on the geometric configuration feature of building, the building contours are extracted by the CSA method, and then the building outlines are derived with regular processing.The method of Fc-S and CSA are tested with Lidar point cloud from Wytheville County, Virginia of America. The results show that the both methods are suitable for mountainous areas, and a higher accuracy can be achived for building and road extraction. Therefore these methods have better practicability. In the mountainous area, Fc algorithm has more efficiency than the Masaharu's algorithm, Fc-S algorithm for building extraction has more accuracy rate than You Hongjian's and LiTao's algorithm, CSA algorithm has more efficiency and details of building outlines than Fc-S algorithm. However, some small buildings are rejected, which may be related to the chosen threshold. This is an insufficient of Fc-S algorithm and CSA algorithm, which needs farther research.