The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B4. Beijing 2008 
316 
Among the available image bands for classification (R, G and B 
from colour images and NIR, R and G bands from CIR images), 
only the bands from CIR images were used due to their better 
resolution and the presence of NIR channel (indispensable for 
grass and trees extraction). In addition, new synthetic bands 
were generated from the selected channels: a) 3 images from 
principal component analysis (PCI, PC2, PC3); b) two images 
from NDVI computation using the NIR-R and R-G channels 
(NDVI1 and NDVI2 respectively) and c) one saturation image 
(called S) obtained by converting the NIR-R-G channels in the 
IHS (Intensity, Hue, Saturation) colour space. 
The separability of the target classes was analyzed through use 
of plots of mean and standard deviation for each class and 
channel and divergence matrix analysis of all possible 
combinations of the three CIR channels and the additional 
channels, mentioned above. The analysis showed that: 
• G and PC2 have high correlation with other bands 
• NIR-R-PC1 is the best combination based on the plot 
analysis 
• NIR band shows good separability based on the divergence 
analysis 
• PC1-NDVI-S combination shows best separability over 
three-band combinations based on the divergence analysis 
Therefore, the combination NIR-R-PC1-NDVI-S was selected 
for classification. The maximum likelihood classification 
method was used. As expected from their low values in the 
divergence matrix, grass and trees, airport buildings and 
residential houses, airport corridors and bare ground, airport 
buildings and bare ground could not be separated. Using the 
height information from nDSM, airport ground and bare ground 
were fused into “ground” and airport buildings with residential 
houses into “buildings”, while trees and grass, as well as 
buildings and ground could be separated. The final 
classification is shown in 
Figure 7. 86% of building class is correctly classified, while 
13% of the reference building data were not detected. Aircrafts 
and vehicles are again mixed with buildings. 
■ Airport 
□ Road 
MTree 
■ Shadow 
ill Grass 
□ Grounc 
Figure 7. Results from supervised classification combined with 
height information. 
4.3 Building extraction using density of DTM-AV and 
NDVI 
Buildings and other objects, like high or dense trees, vehicles, 
aircrafts, etc. are characterized by null or very low density in 
the DTM-AV point cloud. Using the vegetation class from 
NDVI channel as a mask, the areas covered by trees are 
eliminated, while small objects (aircrafts, vehicles) are 
eliminated by deleting them, if their area was smaller than 25m 2 . 
Figure 8 shows the final building class. 87% of building class 
pixels are correctly classified, while 13% of the reference data 
are not detected. 
Figure 8. Lidar DSM points located on buildings, extracted 
using DTM-AV voids and NDVI. 
4.4 Building and tree extraction from Lidar data 
As mentioned above, in DSM raw data the point density is 
generally much higher at trees than at open terrain or buildings. 
On the other hand, tree areas have low density in DTM-AV data, 
as it can be seen in Figure 4. We start from regions that are 
voids or have low density in the raw DTM. These regions 
represent mainly buildings and trees and are used as mask to 
select the raw DSM points for further analysis. 
The first step is identification of buildings by fitting planes, and 
the elimination of these points. The reason is that building roofs 
may cause problems in the subsequent step of vertical point 
density and distribution analysis, aimed at identification of trees. 
The plane fit operation is possible with different commercial 
software. Here, Geomagic Studio by (Geomagic Studio, 
Raindrop Geomagic, Research Triangle Park, NC) was used. 
The planes of small buildings and non-planar parts of large 
buildings could not be detected. With the remaining points, the 
analysis regarding vertical point density and distribution is 
applied in search windows with size 2.5 m x 2.5 m. This size 
was selected larger than the average point density of the DSM. 
From Figure 3, we see that the average density is 1.5-2 points 
for a window of 1.4 m x 1.4 m. 
The points in each search window are projected onto the xz 
plane and divided in four equal subregions, using x min , x max , z min 
and z max as boundary values, with x max = x min + 2.5m. The 
density in the four subregions is computed (see example in 
Figure 9). 
■ 
, m m • 
' m « 
* 
* _ 1 
■ 
' « 
1 
2 
• 
mm mm wy/t 
m 
X 
Figure 9. Projected points in xz plane and 4 subregions.