Contour Clustering Analysis for Building Reconstruction from LIDAR Data

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 our method has following advantages:(1) the closeness of contours can effectively avoid the difficulty of edges grouping in conventional reconstruction methods;(2) using contour cluster analysis can extract different hierarchical structures of the complex building;and(3) even curved surface buildings can be correctly constructed using our method.