The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B5. Beijing 2008 
1088 
Figure 5. Results of the image-based 3D modeling of the Erechtheion in the Acropolis of Athens, shown in wireframe, shaded 
and textured mode. 
Figure 6. The interior room of the Erechtheion 
(above) and part of a wall modeled with Depth from 
Shading (below). 
In the 3D comparison, if the minimization of the distances is 
done on the entire data, the laser’s penetration error will be 
absorbed by the 3D transformation parameters. This is shown in 
Figure 9, with a colour-coded result of the comparison. Clearly, 
the error distribution shows that the largest errors are 
consistently near boundaries, at sharp surface gradients and, in 
addition, at image areas with little texture or repetitive patterns 
(where matching often fails). All these are areas where the 
different modeling methodologies have problems. 
The standard deviation of the differences between the scanned 
model and the image-based model is ca 1.4 mm. As expected, 
the apparent penetration’s effect of the laser into the marble 
surface has been clearly absorbed by the ICP registration. 
For this reason, the best way to register and compare the two 
sets of 3D data is (i) to apply a constraint in the depth direction 
(Z) and compute the shift parameters only in X and Y (Gruen 
and Akca, 2005) or (ii) to match fixed markers or small 
dark/dirt features. The latter method was used to compare 
image-based and range-based results of the area depicted in 
Figure 10. The matching results were compared to the range 
data, co-registering the two surfaces with common features (dirt 
and dark spots), assuming that there is no laser penetration on 
these features. 
A better way would be to use for such test purposes flat, well- 
distributed targets, with good reflectance and no penetration for 
laser scanning and good texture for matching. 
The graphical results of the comparison (Figures 10a and 10b) 
show an apparent penetration of the laser signal into the marble 
(red areas) with an average of ca 3 mm using both Canon and 
Mamiya. Regarding Figure 10, it is not possible to make any 
statement regarding whether image matching or laser scanning 
is more accurate, as better reference data are lacking and are 
difficult in general to get and compare.