The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B5. Beijing 2008 
Figure 3. Results of the data segmentation using the mean-shift algorithm, (top) segmentation of the distance channel, (middle) 
segmentation of the surface normals channel, (bottom) segmentation of the true-color channel. 
Figure 4. Results of the augmented segmentation based on the integration of the three channels. 
3. RESULTS AND DISCUSSION 
The integration of the segments into an augmented 
segmentation is presented in Figure 4. As the Figure shows the 
more dominant segments form the segmentation. The largest 
segment that was selected is the ground segment as derived 
from the normals channel. The second segment is blue façade 
(center building) which was extracted from color channel. Its 
extraction from that channel has to do with the strong and 
almost uniform intensity along it. The third segment is the green 
façade which was derived, again, from the normals 
segmentation. It is of interest to note that the third largest 
segment arrived from the range channel (relating also to that 
façade), but was discarded due to the variability. The fourth 
largest segment is the water fountain in the front of the scan. 
Notice how it did not exhibit any clear structure in neither the 
normals nor the color channels, and was extracted as an 
individual segment from the range channel. In relation to the 
fountain we also note that its border which was not segmented 
as a unique entity in any of the segmentations was extracted as a 
single entity due to the subtraction applied on the extracted 
segments from the individual channels. Additional noteworthy 
elements that were extracted by the augmented segmentation are 
vehicles which mostly were extracted as complete entities, and 
pole elements like the traffic sign at the left of the scan, or the 
streetlamp close to the building façade on the left. 
The halos surrounding the poles at the center of the scan are 
ground points that were not segmented as such, as they were 
lying next to edge points (transition between objects). They 
were correctly separated from the pole in the range channel. 
Similar edge effects can be noticed in other parts like the 
cornice surrounding the blue façade building. Another feature of 
the segmentation is the partitioning of entities due to occlusion. 
This can be noticed in the wall on the leftmost building that was 
partitioned into three large segments. This partitioning is due to 
objects in front of it which hide parts of it. 
749