The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part Bl. Beijing 2008 
1067 
3. 3D CITY MODELLING WITH PICTOMETRY 
DIGITAL IMAGERY 
Oblique images have advantage against vertical images in 
creating building textures since they provide better side view of 
building facades. Pictometry imaging system captures oblique 
images from different directions which are ideal for generating 
building textures. The vertical images taken in the same area 
can be used to generate 3D building models or to refine 3D 
building models created from other data source such as LiDAR 
data. In this section, some issues in generating 3D city models 
using Pictometry digital images will be addressed. 
3.1. Refining 3D Building Models 
3D building models can be extracted from both aerial images 
and LiDAR data. In automatic building extraction from aerial 
images, image matching technique is usually used to extract 3D 
information of buildings. One major problem with automatic 
approaches is that the extraction may fail when occlusions and 
shadows occur in the images. Modem LiDAR systems are 
capable of receiving multiple returns with some penetrating 
vegetation, and thus, the effect of occlusions can be reduced by 
combining the information from different returns, e.g. first and 
last returns and buildings can be extracted reliably. Figure 4 
shows an example of building footprints extracted from LiDAR 
data in downtown area of Buffalo, New York. However, the 
problem with extraction of building models from LiDAR data is 
that the extracted models may not be very accurate because of 
point spacing, scanning angle, the performance of line 
extraction algorithm, etc. Therefore, building models derived 
from LiDAR data need to be refined, in order to create accurate 
3D city models. To correct building models, they are projected 
back on the vertical image triangulated with accurate ground 
control points. The difference between a projected roof edge 
and its corresponding edge extracted from the image is usually 
just a few pixels, as shown in Figure 5. Therefore, an affine 
transformation can be used to correct the building models and 
the transformation parameters can be estimated by using the 
distance between the projected roof edges and the extracted 
edges from the image. 
Figure 4. Building footprints extracted from LiDAR data 
3.2. Selection of Oblique Image 
Due to the image overlap, each building is imaged on several 
oblique images. It is very important to choose one from them to 
give the best texture of the building. In Zebedin et al (2006), a 
score is assigned to all oblique images based on the angle 
between the normal vector of the facade to be textured and the 
vector from the center of the façade to the camera center of an 
oblique image and the one with highest score is chosen. Since 
Pictometry oblique images are captured at a certain angle, a 
reference vector with a certain angle to the building facade 
within the vertical plane passing through the normal vector 
instead of a normal vector is chosen as shown in Figure 6 and a 
Figure 5. Projected building footprint on aerial image 
score is given to an oblique image based on the angle between 
the reference vector and the vector from the center of the façade 
to the camera center of the oblique image. At the same time, a 
visibility analysis is performed to make sure that the façade is 
not blocked by other buildings. 
Figure 6. Selection of oblique image 
3.3. Texturing with Oblique Image 
Once a right oblique image is chosen, the next step is to pick up 
the right image portion and add it to the building façade. To 
make sure the right image portion is selected, it is necessary to 
check whether the building façade projected onto the oblique 
image matches the building edges on the image. The projected 
boundaries of the façade should match the corresponding 
building edges in the image when the oblique image has 
accurate exterior orientation (EO) parameters. However, they 
may not exactly coincide with the actual edges of the building 
as shown in Figure 7, when image's EO parameters from 
GPS/IMU are used directly. To create accurate 3D city models, 
accurate EO parameters of images must be used. The usual way