CMRT09: Object Extraction for 3D City Models, Road Databases and Traffic Monitoring - Concepts, Algorithms, and Evaluation 
and it contains 990 points and 1980 cameras views (see Figure 
1). It includes loops, self-intersections and close parallel roads. 
As a result a building wall can be seen from several locations 
within the path. 
Figure 1. Test area, Rennes historical center. The virtual path is 
depicted in red. 
3. 2D RAY TRACING 
3.1 Principle 
The 2D approach is based on ray-tracing. Each camera is 
analyzed in turn. The walls are represented by 2D segments. 
For each camera a set of compatible wall segments is pre 
selected using three criteria (see Figure 2): 
Distance criterion: the wall is located within a given 
distance from the camera center. 
Half-plane criterion: at least one point of the wall 
segment is located in the half-space in front of the 
camera 
Backface culling criterion: the wall is facing the 
camera. 
The compatible wall segments define a set of candidate walls 
that might be visible from the current camera. An example of 
pre-selection is shown in Figure lOa-b. 
Figure 2. The three criteria for the selection of candidate walls: 
(black=pre-selected walls, red=rejected walls) 
The 2D-tracing technique is then applied to the candidate wall 
segments. First a beam of 2D lines is defined passing through 
the camera center point and regularly distributed within the 
field of view of the camera. Then the closest intersected 
candidate wall segment is selected as a visible wall. When all 
the cameras have been processed then each wall can be 
associated to the list of cameras that can view it. 
3.2 Test results 
The method was tested with various numbers of rays per 
camera. The distance threshold was arbitrarily set to 150m, 
distance above which the texture resolution is low enough to be 
discarded. The computing time includes reading and exporting 
steps. Numerical results are shown in Table 1. Between 10 and 
13% of the walls are detected as visible by the process. Figure 3 
shows the evolution of the wall number and the computing time 
with the number of rays. A qualitative example of selected 
walls can be found in Figure 10c. 
Ray # 
Total # of 
selected walls 
Avg # of cameras 
per wall 
Computing 
time 
10 
1176(10.3%) 
4.54 
7s 
50 
1391 (12.1%) 
4.95 
11s 
100 
1450(12.7%) 
5.04 
17s 
500 
1507 (13.2%) 
5.14 
50s 
Table 1. Results of 2D ray tracing 
0 100 200 300 400 500 600 
Rays # 
0) 
E 
"■3 
U) 
c 
V- 
3 
Q. 
E 
o 
o 
o 
3 
CD 
Figure 3 -Number of visible walls and computing time in 
relation to the number of rays 
3.3 Discussion 
The variations in the number of selected walls come either from 
walls located far away of the camera, or from walls almost 
aligned with the camera center. When the number of rays is 
small, then many walls are located between two rays and are 
therefore not selected (see Figure 4). In our configuration, a 
number of rays around 100 seems to be a good compromise to 
get a maximum number of relevant images per building wall. 
The main advantage of the 2D approach is the speed. It is also 
very simple and quick to implement. However it does not take 
building heights into account. Yet a low building (garage, shop, 
etc) may only mask the bottom part of higher buildings located 
behind it, especially if the camera is located on the top of a 
vehicle (see examples in Figure 5 and Figure 11 a-c). Therefore 
it seems very important to make use of 3D information within 
the selection process.