In: Paparoditis N., Pierrot-Deseilligny M.. Mallet C.. Tournaire O. (Eds), IAPRS. Vol. XXXVIII, Part ЗА - Saint-Mandé, France. September 1-3, 2010 
296 
the X direction is the road segment direction. The half-width is 
defined as follows: 
7) — 1 
half Width = V ■ Wt, ( 1 ) 
?=l 
where: 
• n is the number of curbstone points for the current side, or 
dered and numbered from 0 to ?? — 1; 
• W% is a weight given by: 
where L is the road segment length. 
The average road segment width is the sum of the left half-width 
and the right half-width. 
4 EXPERIMENTATION AND RESULTS 
The algorithm was tested on a data set acquired in Cavell street, 
a narrow street in the historical town centre of Rennes, in France. 
This street is characterized by a significant slope and the pres 
ence of several parking lots and parked cars (see Figure 4). Both 
the left and the right laser point clouds were available on this 
street, but not on perpendicular streets. Figure 5 shows both laser 
point clouds used for this study. There are many cars occluding 
the curbstones on the right side. The results at each step of the 
method, as well as the final result, are presented in this section. 
Figure 4: Real view of Cavell street. 
4.1 Ground points extraction 
First, the ground laser points is extracted from both the right 
and the left point clouds, using the method presented in section 
3.1. Figure 6 shows the resulting ground points. Despite the 
street slope, the ground points are correctly extracted all along 
the street. 
4.2 Curbstone point detection 
Curbstones are then independently extracted from the ground point 
clouds located on the left and on the right sides of the vehicle, fol 
lowing the method proposed in section 3.2. Figure 7 shows the 
gradient images associated to the two point clouds. The pixel size 
is about 13.3 centimetres. These images are binarized in order to 
provide pavement pixels and corresponding laser points. 
Figure 5: Top and side views of the left (light orange) and the 
right (dark blue) original point clouds from Cavell street. 
Figure 6: Top and side views of the ground points extracted on 
Cavell street. The second one shows the street slope. 
The curbstone points are then derived from these images, as it 
is shown in Figure 8. The right laser point cloud contains many 
parked cars occluding the pavement edges. Some curbstone points 
are detected between cars, that is sufficient to recover pavement 
edges despite occlusions. 
4.3 Pavement edge ordering and connection 
After curbstone points are detected, they are ordered and con 
nected into pavement edges (cf. section 3.3). The result of this 
step on our test data is presented in Figure 9. The connected com 
ponents have been linked when they were near enough but the 
lack of detected curbstone points on the right cloud has produced 
a break in the right pavement edge.