基于等高线簇分析的复杂建筑物模型重建方法  

Contour Clustering Analysis for Building Reconstruction from LIDAR Data

作　　者：张靖[1] 李乐林[1] 江万寿[1] 

机构地区：[1]武汉大学测绘遥感信息工程国家重点实验室,武汉430079

出　　处：《地球信息科学学报》2010年第5期641-648,共8页Journal of Geo-information Science

基　　金：国家自然科学基金项目"基于等高线族分析的LIDAR数据建筑物提取研究"(40671159);"863"课题"基于物探飞行模式的多航空遥感传感器集成系统研制"(2006AA06A208)

摘　　要：近年来,基于LIDAR点云数据的建筑物重建模型一直是研究的热点。目前,出现的许多算法对简单建筑物,如平顶房屋、人字行屋顶及其他规则房屋的重建取得了不错的效果,但是,对于结构复杂的建筑物重建问题仍然有待解决。针对这一问题,本文提出了一种利用等高线簇分析从LIDAR数据中自动重建复杂建筑物模型的新算法。该算法是一种自底向上的数据驱动方法,以等高线所反映出的建筑物轮廓特征为基础,充分利用等高线封闭性和明确的拓扑关系,采用等高线形状分析的方法来实现建筑物的检测和模型识别与重建。算法实现分为4个步骤,首先,通过对LIDAR点云数据的DELAUNAY三角化跟踪提取等高线,然后利用等高线的长度,面积等形状参数来提取建筑物等高线,再通过拓扑分析,以及形状匹配的方法对等高线进行分簇,得到同一建筑物不同组成部分的等高线簇,最后,对各簇等高线进行模型参数优化并按拓扑关系进行重组得到完整的建筑物模型。通过对多层次、多曲面等复杂建筑物的重建实验证明了此方法的可行性。The automatic building reconstruction from LIDAR data has been a hot issue for several years.Many methods and algorithms have been put forward to reconstruct the models of simple buildings,such as of flat roof,gable roof,or other rectangular shape.But it remains an open problem for the reconstruction of complex buildings.In this paper,a new idea is introduced to process the complex buildings.It is based on the contour clustering analysis of LADAR data and a bottom-up method using data-driven processing.Contours contain the shape information of object boundary,and they are closed,complete,and having explicit topological relationships among each other.So we can use these valuable characters to guide the building reconstruction.In this paper,a concept of contours cluster is introduced,which is based on the following observations： the contours of a building are usually very similar to each other in every part of the building;the nested similar contours are defined as a cluster of similar contour.The contour cluster reflects the detailed feature of the corresponding object.Analyzing the shape differences among contours clustering,the different parts of the whole complex building can be found out.So we can say,contour clustering is very useful,and is the core of the method.The process includes 4 main steps.Firstly,the LIDAR point is pre-processed,and the Delaunay mesh is constructed with the processed LIDAR points and the initial contours are traced.Secondly,some shape features are used to distinguish the contours on buildings or on other objects.With thresholds of contour length and area,some contours of vegetation can be removed.Thirdly the topology relationship and similarity relationship between contours are analyzed.Based on these relations,the contours are clustered to form the parts of buildings.At last,the building model of different types can be reconstructed from the clusters of contours.To test the approach presented above,2 experiment data with representative building models are applied.The results show o

关 键 词：LIDAR 等高线 分簇 形状匹配 

分 类 号：P237[天文地球—摄影测量与遥感]