688 
The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B5. Beijing 2008 
transforms the building façade to fit within the image according 
to the image’s extrinsic and intrinsic parameters. The forward 
mapping method is a more tedious process as the user is 
required to provide the outline of each façade on the image 
before texture mapping can be carried out, hence this method is 
not suitable for our application. On the other hand, the reverse 
mapping method has a greater potential for workflow 
optimization as the façade outlines can be projected 
automatically onto the image, provided the full set of camera 
parameters are known. The reverse mapping method will be 
used as the basis for texture mapping of building façade in our 
system. 
2.1 Camera calibration and image orientation recovery 
The projection of the building model façade onto the close- 
range digital image is done using collinearity equations 
(McGlone, 2004). In order to project the façade accurately, it is 
important to obtain the extrinsic and intrinsic parameters of the 
digital camera. 
The intrinsic parameters of the camera comprises of the pixel 
size, focal length, image size and lens distortion. The focal 
length and image size is readily available in the image’s 
Exchangeable Image File Format (EXIF) while the sensor size 
information can be found in the specific camera manufacturer’s 
specifications. Lens distortions are assumed to be negligible due 
to the distances of which the target object is away from the 
exposure station. 
The extrinsic parameters of the camera are the metric position 
and orientation when the image was taken with respect to the 
world reference frame. The extrinsic parameters can be 
recovered with a combination of a closed form space resection 
(Zeng and Wang, 1992) and classical non-linear space resection 
based on collinearity (McGlone, 2004; Wolf, 2000) using just 4 
control points to match object space to image space. The closed 
form space resection approximate solution provides an initial 
estimate for the classical non-linear space resection method to 
solve for the precise extrinsic parameters. 
2.2 Culling of non-front facing and occluded façade 
As the reverse mapping method of projecting building model 
coordinates onto the image is used, non-front facing façade will 
be projected onto the image as well. This is not acceptable as 
the non-front facing façade will cause erroneous texture 
mapping of the building models. To prevent this, we will have 
to decide whether the projected façade is front facing or non- 
front facing before texturing can be done. The building façade 
world coordinate points are arranged in a clockwise manner to 
denote a front facing façade as shown in Figure 1. After 
projection onto the image, the winding of a front facing façade 
should still be clockwise while a non-front facing façade will 
possess an anti-clockwise winding, thus effectively culling all 
non-front facing façade. 
It is our intent to carry out large scale texturing for the building 
façade, hence we try to capture as many buildings possible on a 
single image. This can be done by moving the exposure station 
as far away possible and keeping an oblique angle to the 
intended buildings to be textured. In the process, it is inevitable 
that complete occlusion of some buildings might take place, 
resulting in the erroneous texturing of those occluded building 
façade. A simple polygon collision test (Miller, 2000) was 
devised and all front-facing façade were tested. If 2 different 
façade are fully overlapped, the façade nearer to the exposure 
station will be chosen to be textured. 
Anti-clockwise 
winding non-front 
facing façade 
Figure 1. Differentiating between front facing and non-front 
facing facade 
2.3 Geometric correction of skewed façade textures 
Oblique view images of the buildings are more effective in 
large scale texture mapping as compared to frontal planar view 
images. It is because they are able to capture more views of the 
building and reduce the number of photographs required. 
However, a skew factor will be introduced to the mapped 
façade after being projected from the building model. The 
resulting texture polygon outline is irregularly shaped and the 
irregularities are highly dependent on the angle of which the 
image was acquired. 
Figure 2. Adjustment of skewed facade texture 
We devised a method in OpenGL to correct the geometric 
distortion due the irregularly shaped polygon. As shown in 
Figure 2, the skewed façade texture is mapped onto an 
intermediate rectangular mesh with sufficient resolution using 
the façade position coordinates (a, b, c & d) to obtain the 
appropriate comer texture coordinates (u a , v a , u b , v b , u c , v c , u d , 
v d ) for the mesh. The other points within the mesh are assigned 
texture coordinates through linear interpolation of the 4 comer 
texture coordinates. The mesh is subsequently rendered via 
OpenGL and the resulting frame buffer is a geometrically 
corrected façade texture. 
2.4 Colour balancing of extracted façade textures 
After texture mapping is carried out, the resulting building 
model would consist of a mosaic of different images acquired 
possibly under different illumination and physical conditions.