文章基本信息

标题：A Novel Method for Extracting Building from LIDAR Data--Fc-S method
本地全文：下载
作者：REN Zizhen ; ZHOU Guoqing ; CEN Minyi 等
期刊名称：ISPRS Annals of the Photogrammetry, Remote Sensing and Spatial Information Sciences
印刷版ISSN：2194-9042
电子版ISSN：2194-9050
出版年度：2008
卷号：XXXVII Part B1
页码：283-288
出版社：Copernicus Publications
摘要：Automated techniques and tools for data acquisition of building are urgently needed because building information is extremely important for many applications such as urban planning, telecommunication, or environment monitoring etc. For the traditional manually building extraction from raw imagery is highly labor-intensive, time-consuming and very expensive, the generation of Three-dimensional building models from point clouds provided by airborne laser scanning, also known as LIDAR (Light Detection And Ranging), is gaining importance. Several methods have been presented for building extraction from LIDAR data during last decades. Based on data source involved, those methods can be divided into two classes; the one is only using LIDAR data and general knowledge about buildings, geometric characteristics such as size, height and shape information to separate buildings from other objects. Another is using LIDAR data fashioned another source data, such as aerial imagery, multispectral images, and so on, to extract building. For the acquisition of multi-source data and the fusion of them still have some difficulties, more researchers pay attention to the former class. This article presented a novel method for extracting building solely based on LIDAR data. For the proposed method consists mainly of two processes: filtering and segment, and in the first process, we used the properties of contours of digital surface models to distinguish the cloud of LIDAR points into on-terrain points and off-terrain points which is different from previous filtering method such as mathematical morphological filter, we named this method as Fc-S method. The main purpose of filtering process in Fc-S method is to generate DTM from LIDAR data. There are four steps: Firstly, digital surface model is generated from original LIDAR point data by a nearest neighbor interpolation method which can preserve the sharp difference between buildings and their surrounding ground. Secondly, contours are derived from digital surface models and are separated into on-terrain contours and off-terrain contours according to some properties of contours, including its closing, distance from the starting point to the last one, length of contour. Thirdly, an initial DTM is generated by interpolating on-terrain contours and refined by iterative method. And lastly, the normalized digital surface models, which can describe those objects(buildings, trees, etc), is generated after the ground information and low objects, e.g. cars, are removed by the height difference between DSM and DTM. To ensure there are no on-terrain points in the normalized digital surface models, the edge information is considered. Those areas with less edge information are delete because building area should has more edge information. Edge information can be easily obtained by sobel edge extraction method. In the second process, the main purpose is to distinguish buildings from vegetation. Because we extract buildings solely from DSM, the criteria must be geometric ones. Based on general knowledge about buildings and vegetation, we try to use size and shape characteristics to achieve this purpose. In the previous segment methods, the size threshold usually is used firstly to remove some smaller objects such as single trees. Thus a tough problem is left: the larger vegetation area or vegetation mixed with buildings cannot be removed using size criterion. In our method, we try to use shape information to reduce the size of large vegetation or separate vegetation from buildings before using size threshold. Observing the building and the vegetation, we can discover a fact that the shape of their top is very different; the one is roughly smooth whether its roof is lever or oblique, and the one is rather undulate. So we select quadratic gradient parameter to distinguish them. After deleting the points which quadratic gradient is bigger than the threshold, the vegetation area remaining become smaller and can be removed by size threshold almost. At last, the fine building is extracted by iteration technique. The article describe the algorithms involved, giving examples for a test site in Fairfield, and the extracted building by the proposed method are evaluated by comparing with the high-resolution aerial photograph acquired in the same area. The results show that the proposed method can extract building relatively accurate in the region with undulatory surface and even can extract building which is adjacent to vegetation
关键词：LIDAR; Building Extracting; Filtering; Segment; DTM