In: Paparoditis N., Pierrot-Deseilligny M„ Mallet C., Tournaire O. (Eds), IAPRS, Vol. XXXVIII, Part 3A - Saint-Mandé, France, September 1-3, 2010 
L 
2 
■■ 
(a) Gradient image derived from the left ground point cloud. 
(b) Gradient image derived from the right ground point cloud. 
Figure 7: Gradient images derived from the two point clouds. 
High gradients are yellow and red. 
Figure 8: Detected curbstone points (black balls) superimposed 
with ground points. 
Figure 9: Connected pavement edges. 
Note that the computed pavement edge can be locally irregular. 
This is due to the width of the line derived from the gradient 
image. For example, it is about two pixels wide for the gradient 
image of Figure 7(a). 
4.4 Road surface modelling 
The laser point clouds obtained with the mobile mapping vehicle 
are compared to the road and building data from the 3D database 
(see Figure 10(a)). The building façades are first registered to the 
laser point clouds and the registration parameters are applied to 
the road segments (see Figure 10(b)). 
The pavement edges from the left and the right clouds are then as 
signed to the nearest road segments to delineate the road surface. 
The final result of this experimentation is presented in Figure 11, 
that shows that the street is correctly delineated. 
4.5 Average road width 
The average width is computed for every processed road segment. 
The green segment of Figure 11 (the northest one) has an aver 
age width of 7.68 meters whereas the red segment (the southest 
one) has an average width of 7.01 meters. The green segment is 
wider because it contains more parking lots than the red one, as 
it can be clearly seen in Figure 11 (a). Indeed, it is the roadway 
width that is estimated rather than the lane width. These results 
are consistent with manual image-based measurements. Manual 
(a) before registration. 
Figure 10: 2D view of road segments, façades and ground points. 
! 
k L 
TT 
.'••// / i 11 
Figure 11: Top and side views of final road surface delineation. 
measurements show an average value of 7.6 meters for the green 
segment and 7.0 meters for the red one, which shows an accuracy 
of a few centimetres. 
4.6 Other results 
The results of the pavement edge splitting step are illustrated on 
another data set of the same type (cf. section 3.4). Two pavement 
edge components are detected on the right cloud of the street 
whereas it is composed of four road segments in the model, as 
it can be seen in Figure 12(a). Figure 12(b) shows the result of 
the splitting step. The blue edge has been divided into four parts 
and each part has been associated to the right road segment. The 
part associated to the red segment has been rightly merged to the 
neighbouring pavement edge. 
297