BUILDING ROOF RECONSTRUCTION BY FUSING LASER RANGE DATA AND 
AERIAL IMAGES 
J.J. Jaw *,C.C. Cheng 
Department of Civil Engineering, National Taiwan University, 1, Roosevelt Rd., Sec. 4, Taipei 10617, 
Taiwan, China - (jejaw,d95521009)@ntu.edu.tw 
Commission III, ThS-7 
KEY WORDS: Topological Relationship, Line Features, Geometric Inference, Data Fusion, Building Roof Reconstruction. 
ABSTRACT: 
The objective of this study is to present an efficient and robust method of building roof reconstruction by fusing laser range data and 
aerial images through CSR (Construct-Shape-Refine) procedures. The algorithm starts by extracting 3D line features from laser 
range data by using the semi-automatic 3D line feature extraction engine. After 3D line features are extracted and regarded as input 
data for the CSR algorithm, then the procedures of building roof reconstruction are performed in the following algorithms of 
geometric inferences: (1) constructing the topological relationship of 3D line features that belong to the same building roof by using 
the special intersecting property of 3D line features projected onto plane; (2) shaping the initial building roof by means of adjusting 
the 3D line features, and compensating missing parts, if any, by the shortest path algorithm and reporting whether or not the 
investigated building roof is completed; (3) as a final stage, refining the building roof automatically or semi-automatically by 
integrating 2D line features observed from the images through geometric inference processes. The experiments show that the 
proposed CSR algorithm provides a workable platform for building roof reconstruction by fusing laser range data and aerial images. 
1. INTRODUCTION 
3D reconstruction of city model has recently been a popular 
research topic in Digital Photogrammetry (DP) as well as 
Computer Vision (CV) community. The indispensable process 
of 3D city modeling is to construct building models. 
Traditionally, the generation of building models is mainly 
performed by measuring the conjugate points on aerial stereo 
pair images. However, line features, of higher-order 
information and easier detected than point features, are the main 
evidence for building hypotheses and an excellent feature 
primitive for building reconstruction if gone through proper 
photogrammetric approaches (Schenk and Csatho, 2002). As far 
as the data sources are concerned, line features can be extracted 
or measured from aerial images, topographic maps, and laser 
range data (also termed LIDAR point clouds or LIDAR data in 
this study), etc. LIDAR system, among which, has emerged as a 
new technology in the past decade for obtaining the surface data 
potentially revealing detailed scene geometry, which, as 
compared to aerial images that contain abundant spectral 
information and scene information, renders an promising 
alternative for feature extraction and building reconstruction. 
Besides, vertical component accuracy is far better than 
horizontal component in airborne LIDAR system while 
photogrammetric means usually suggests the opposite result due 
to restricted base/height geometry. It is therefore found that the 
LIDAR point clouds and aerial imagery data possess mutually 
independent advantages, suggesting a complementary potential 
if they are appropriately fused. Rottensteiner and Jansa (2002) 
proposed a bottom-up algorithm of fusing LIDAR data with 
aerial images for handling polyhedral buildings of arbitrary 
shape without any prior information about the building 
outlines. Seo (2003) presented an integration method based on 
fusing LIDAR data with aerial images to increase the level of 
automation in building recognition and reconstruction. Ma 
(2004) proposed a scheme of building reconstruction by fusing 
LIDAR data sets with aerial images based on polyhedral model, 
where aerial image data are meant to improve the geometric 
accuracy of the building model. Chen et al. (2006) proposed 
fusing LIDAR data with aerial images to detect building regions 
followed by a reconstruction strategy of Split-Merge-Shape 
processing. 
LIDAR 
3D line feature 
extraction 
i'frr.m TTnAR'l 
Topology 
Initial roof model 
construction 
(LIDAR system) 
Corresponding author. 
Input Data 
Image feature 
extraction 
(Point or Line)