The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B3b. Beijing 2008 
603 
Figure 6: a) Original image b) Moving objects c) Mixture of 
Gaussians mask d) Bayes algorithm mask e) Lluis-Miralles- 
Bastidas mask f) Non-parametric model mask. 
execution time per frame 
130 
n 
595 
726 
295 
70 
I—I 
386 
194 
All 
275 
54 
1 1 
353 
377 
0 0(0 0 0 
Tî- CO h- T}- CO 
CNjTiinCNTj- 
X X X X X 
O O CO o o 
(NlTtCOCNTt 
CO CO f". CO CO 
CO o o 
■O- CO 
m CN Tt 
XXX 
00 o o 
CO CN Tt 
I s - CO CO 
768x576 
320x240 
640x480 
768x576 
Bayes MoG Lluis Non parametric 
In order to analyse the results acquired by the system, ground 
truth data had to be made available. For the airport prototype, 
tests have been conducted using cars equipped with GPS 
devices. For the tunnel prototype on the other hand, the ground 
truth data was collected by viewing image sequences through a 
specially developed manual marking tool (Figure 8), which 
allows the user to specify the correct position of the moving 
targets. The ground truth data has been correlated with the 
system results and then inserted into a database, along with 
other information such as weather conditions. A test analysis 
tool has also been implemented in order to display various 
statistics that can be acquired by querying the database. 
This test analysis tool was used in order to provide qualitative 
comparison of the two operation modes of the system, the grid 
and map fusion modes. A very crucial statistic that is suitable 
for this kind of evaluation is the absence error. Its significance 
lays on the fact that a high value of this statistic means that the 
system is prone to be lead to wrong conclusions with severe 
consequences, such as failing to identify an emergency situation. 
As it can be seen on Figure 9 the absence error appears more 
rarely than the other two types of errors (presence and position) 
for both modes. Especially in the map fusion mode, this statistic 
is even lower, achieving half the value of the one acquired from 
the grid mode. 
Although the map fusion method appears to provide more 
accurate results, there are other areas where the grid mode 
shows better performance, such as the utilisation of less 
bandwidth. The volume of the data transmitted by each ATU at 
every frame circle for grid fusion as measured for the prototype 
applications was 1 Kbyte/frame while for the map fusion mode 
it was measured at 8Kbyte/frame. 
Another characteristic of the two methods that is worth 
mentioning is the execution times they achieve, both for the 
ATUs and the SDF. As shown in Figure 10, the grid mode is 
more intensive for the ATUs while the map fusion mode evokes 
bigger execution times at the SDF server. In other words, the 
choice of operation mode is a complex decision that should be 
based on several factors, such as the network capabilities, the 
available computation power of the SDF server unit and the 
total number of ATUs used on the particular configuration of 
the system. 
Figure 7: Background extraction methods execution times 
error distribution per mode 
grid mode map fusion mode 
Figure 9: Percentage of error distribution per mode 
Figure 8: Screenshot from the Manual Marking tool