Figure 3. Result for image sequence Three Walls with 36 images, no approximations for position and attitude given, o0= 0.67 pixels. 
( 
Figure 2. Image sequence Three Walls, images 3, 9, 12, 20, 25, 
and 28 - Please note the dark approximately horizontal line 
which stems from the ring around the UAV to secure its rotors 
against hitting other objects. 
The experiments were conducted under challenging wind 
condi-tions. Due to the very light UAV it was difficult to fly to a 
position - already slight gusts of wind had a strong effect on the 
UAV. The images were thus taken only very roughly on a 
circular path, they had a rather variable orientation (cf. Figure 2). 
and several of them were not in the correct order. As the order 
in the sequence is a basic requirement of our approach for 3D 
reconstruction, we had to organize the images into a continuous 
sequence with threefold overlap by hand. 
A sequence with 36 images could be found for which a 3D 
model could be reconstructed (cf. Figure 3) showing Three 
Walls. Figure 4 presents the reconstructed planes also together 
with the projected texture computed by means of consensus (cf. 
Section 2).