CMRT09: Object Extraction for 3D City Models, Road Databases and Traffic Monitoring - Concepts, Algorithms, and Evaluation 
LIDAR road points are incorporated to provide height 
information for three-dimensional surface modelling. A 
threshold for the maximum elevation histogram is used to 
determine which method will be used for road surface 
initialization, i.e., either surface fitting or lowest point selection. 
The radius of a buffer circle is set to be half the roadwidth. 
Based on the experience, this percentage threshold of elevation 
histogram of initial surface modelling would be 30%. The 
reason is that the space interval of the along-track vertices is 
densified to 0.5m. Those local slopes of the vertical profiles are 
assumed to be less than 45 , i.e., the elevation change is smaller 
than 1,5m. Since the slope of a road is seldom larger than 45 , 
the threshold is reasonable that adapts for general applications. 
Possible interference could be the presence of dense vehicles 
that make the point clouds deviate from the road surface. This 
could lead to unreliable results. For more precise surface 
modelling, the spanning distance between pseudo nodes is set to 
be 200 m, according to the rules of roadway designs. Next, the 
vertical profiles, cross-sections, and intersections are smoothed 
according to either height difference or iterative times. Figure 5 
shows the reconstructed three-dimensional road models. 
To evaluate the reconstructed road models, reference LIDAR 
road points are extracted manually. The normal height 
differences between the reference points and the reconstructed 
surfaces are compared to calculate the relative error assessment. 
The index of modelling error is expressed as the root mean 
square error (RMSE). The generated results indicate that the 
RMSEs for the modelled surfaces of test sites are lower than 
0.15 m. Those values indicate that the iteratively local approach 
may lead to modelling errors within the range of random error 
of the raw data. The slopes of reconstructed models in vertical 
profiles and cross-sections are estimated and shown in Figure 6 
and 7, respectively. 
Ml 
Figure 4. Digitized road boundaries in aerial images 
ÊT 
] <s> 
Figure 2. Road boundaries in topographic maps 
ÏÎ 
1+20.51 
1+19.05 
1+17.58 
J+16.12 
¡+14.65 
1+13.19 
1+11.72 
1+10.26 
1+879 
1+7.33 
1+5 88 (unit: m) 
Figure 3. Sub part of laser scanning data 
Figure 5. One part of reconstructed models (Overpass) 
0 2 4 6 8 10 12 14 16 18 20 
Figure 6. Histogram of slope and slope difference of along-track 
profiles 
10