In: Wagner W., Székely. B. (eds.): ISPRS TC VII Symposium - 100 Years ISPRS, Vienna, Austria, July 5-7, 2010, IAPRS, Vol. XXXVIII, Part 7B 
CHANGE DETECTION OF BUILDING MODELS 
FROM AERIAL IMAGES AND LIDAR DATA 
Liang-Chien Chen a ’*, Li-Jer Lin b , Hong-Kuei Cheng, c Shin-Hui Lee, c 
a Center for Space and Remote Sensing Research, National Central University, No.300, Jhongda Rd., Jhongli City, 
Taoyuan County 320, Taiwan - lcchen@csrsr.ncu.edu.tw 
b Department of Civil Engineering, National Central University, No.300, Jhongda Rd., Jhongli City, Taoyuan County 
320, Taiwan - 973202088@cc.ncu.edu.tw 
c CECI Engineering Consultants, Inc., 28th Floor, No. 185, Sec. 2, Sinhai Rd., Taipei City 10637, Taiwan - (tc561, 
shl)@ceci.com.tw 
KEY WORDS: Change detection, Building, LIDAR, Image, Aerial 
ABSTRACT: 
Building models are built to provide three dimensional (3D) spatial information, which is needed for varieties of applications, such 
as city planning, construction of location-based services, and the like. However, three dimensional building models need to be 
updated from time to time. Rather than reconstructing building models for the entire area, it would be more effective to only revise 
the parts that have changed. In this study, we aim at finding changes with 3D building models. The proposed scheme comprises three 
steps, namely, (1) data registration, (2) change detection of three dimensional building models, and (3) detection of new building 
models. The first step performs data registration for multi-source data. The second step performs the rule-based change detection, it 
include examination of spectrum from aerial images, examination of height difference between building models and LIDAR points, 
and examination of linear features from aerial images. A double-threshold strategy is applied to cope with the highly sensitive 
thresholding often encountered when using the rule-based approach. In the third step, we detect the LIDAR point clouds in the new 
building areas by removing vegetation, ground and old building areas. We then use region growing to separate the LIDAR point 
clouds into different groups. Finally, we use boundary tracing to get the new building areas. Ground truth data are used for validation. 
The experimental results indicate that the double-threshold strategy improves the overall accuracy from 93.1% to 95.9%. To provide 
comprehensive observations, the different cases are scrutinized. 
1. INTRODUCTION 
A cyber city can be constructed which contains more spatial 
information than traditional two-dimensional topographic maps. 
This also provides the possibility to comprehensively integrate 
various types of 3D information. Three dimensional building 
models are one important part of a cyber city. Considering the 
rapidity of urban growth, a 3D geographic system is in need for 
updating the building models in the 3D information system. The 
effective revision of spatial data becomes important. Currently, 
change detection is usually done through spectral analysis of 
multi-temporal images. Nevertheless, building models also have 
three-dimensional information. So, we try to fuse the LIDAR 
data and aerial images for building model change detection. 
LIDAR data and aerial images have their own particular 
advantages and disadvantages in terms of horizontal and 
vertical accuracy. Compared with aerial images, LIDAR data 
provide more accurate height information but less accurate 
boundaries. Aerial images provide more extensive 2D 
information such as high resolution texture and color 
information. Although 3D height information can be estimated 
from one or several images by the use of several methods (such 
as stereo, shape from shading, comparison to LIDAR) the 
height information extracted from aerial images is still relatively 
less accurate. (Lee et al., 2008). 
Several studies of change detection using spectral imagery have 
been reported (Mettemicht, 1999). Recently, a number of 
change detection methods using LIDAR data have been 
proposed. Murakami et al. (1999) used multi-temporal LIDAR 
data to produce Digital Surface Models (DSMs) for the 
detection of changes. Walter (2004) used LIDAR data for 
object-based classification and observation of land phenomena 
to determine the land-use category. There has been many 
studies using the vector maps (Knudsen and Olsen, 2003; 
Matikainen et al., 2004), LIDAR data (Girardeau-Montau et al., 
2005; Murakami et al., 1999), and aerial imagery (Jung, 2004) 
as the old data set. Some have used 3D building models as the 
old data set for this purpose (Huang, 2008). 
2. METHODOLOGY 
Since Lidar data and aerial images have unique advantages and 
disadvantages, it is natural to integrate those two data sets. In 
this paper, we aim to find the changed 3D building models 
using old building models with new LIDAR data and aerial 
imagery. It includes two parts in change detection. One is 
change detection of old buildings; the other is detection of new 
buildings. The workflow is shown in Fig 1. 
Corresponding author.