532 
The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B3b. Beijing 2008 
pixel and satisfies the following condition, 0 < N{p) < 8, where 
N (p) is the number of nonzero neighbors of pixel p, i.e., 
7 
N(p) = Y,Pi (13) 
/=0 
In the designed vehicle extraction approach, we try to use the 
uniform of radiations from road. In this case, the road cluster is 
selected and the vehicle in the segmented imagery can be seen 
as “noise”. Figure 5 shows the binary road imagery. 
P7 
Pii 
Pi 
P 
P2 
PS 
P4 
P3 
Figure 2. Neighbourhoods arrangement 
Figure 5. Binary Road Imagery 
Above road imagery is filtered by binary morphological open 
operation to obtain solid road. In this experiment, the 
structuring element is set as 5. The result of the morphological 
operation is in Figure 6. 
4. EXPERIMENTS AND RESULTS 
The proposed road extraction algorithm has been tested on 
UAV aerial imagery, Figure 3 shows two testing images. 
Figure 3. Testing Imagery. 
Figure 6. Filtering Binary Road Imagery 
In order to extract the vehicles from imagery in Figure 5, we 
subtract the imagery in Figure 5 from the imagery in Figure 6. 
Figure 7 demonstrates the vehicles extracted by the subtraction 
operation. 
In colour segmentation operation, the parameters k\ and k 2 in 
Equation (5) for calculation of the colour similarities are chosen 
to be 0.0001 and 0.2, respectively. The number of the clusters is 
set as 4 and the maximum iterates is 50. The crossover and 
mutation probability are 20%. Figure 4 gives the results of the 
colour segmentation. 
Figure 7. Binary Vehicle Imagery 
The binary morphological operations are carried on the vehicle 
imagery to obtain complete vehicles as shown in Figure 8. 
Figure 4. Segmented Imagery