The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B3b. Beijing 2008 
Figure 2: Illumination variation before failure in data set 1 (up) 
and data set 2 (down). The left figures are the reference images 
and the right ones are the candidate images. The difference 
between the images is the transformation of the whole image 
and illumination differences. 
5. CONCLUSION 
We have evaluated and quantified the robustness of our 
stabilization method with respect to the amount of disturbances. 
The disturbances were simulated as either moving objects or 
illumination variation. The acceptance of our method is decided 
based on acceptable parameter and geometric errors. 
The simulation is done based on extreme case of illumination 
variation and moving objects. The presented percentage of 
amount of the disturbances can be increased in the case of 
having illumination variations with non destructive nature or 
moving object with low motion correlation. 
More disturbances would be handled by having more structure 
in the image. The results demonstrate the low percentage of 
disturbances in the acceptance boundaries in an image with 
almost no structure outside the road area and having a road in 
the middle of the image. However the stabilization of this case 
can be done without having a very high amount of disturbances. 
This data set is an extreme case. In the other data set with 
highway crossing, even a very high amount of the disturbances 
in both moving objects and illumination variations provide 
acceptable results. This case shows the robustness of our 
method in handling very large disturbances which in reality 
would not occur. 
disturbances % 
disturbances % 
5 5 
—data! 
—<Sata2 
0 
5 
•j 
.? 
i 
disturbances % 
Figure 3: Moving object boundaries 
Data set 1 is represented in blue and data set 2 in red. The 
acceptable percentage of the amount of the illumination 
variation is represented by the star. The region between the star 
and rectangle is the optimizer failure within our settings. The 
circle shows one of the failure cases from this type. The area 
after rectangle shows the real rejection. The method is robust 
before the specific amount of moving objects which is indicated 
by star. All the normalized parameter errors are also zero.