459 
The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B4. Beijing 2008 
larger radial distances between a DSM cell and the nadir point 
can be also helpful in detecting the points producing occlusions. 
By combining both of the above methods, we can enhance the 
capability of our procedure to detect the points causing 
occlusions. 
PC 
Figure 2. Occlusion detection in perspective views. 
Figure 3, the building requires two perspective centers on 
opposite sides in order to detect all the non-ground points for 
each vertical profile. It is necessary to check all possible 
occluding directions. Considering that every pixel has eight 
possible neighbors that could produce occlusions, for each pixel, 
we use 8 perspective centers with heights close to the maximum 
elevation of the entire area; this way, the points causing 
occlusions can be detected more thoroughly. For the same 
reason, larger radial distances between DSM cells and the nadir 
points are also required. The locations of the perspective centers 
are outside the region of interest, at a distance d. 
The algorithm is tested using the artificial data with sloping 
terrain shown in Figure 4. Some objects are located above the 
ground, and some noise is added to the DSM. Using the 
proposed algorithm with synthesized perspective centers, 
potential non-ground points are separated from ground points, 
as shown in Figure 5, in which the white points are the points 
causing occlusions and the black points are the extracted ground 
points. The result of classification in Figure 5 includes some 
false hypotheses which are caused by noise in the terrain surface. 
1 
I 
I 
i 
& 
V 
V 
/ 
/ 
/ 
A 
/*■ 
Figure 3: Two perspective centers on opposite sides used to 
detect all the non-ground points 
When dealing with large buildings, detecting non-ground points 
using only one perspective center could be a challenge. In 
Figure 4: The simulated surface Model. 
Figure 5: Points causing occlusions.