963 
Institute of Geodesy and Photogrammetry, Swiss Federal Institute of Technology (ETH) Zurich, ETH-Hoenggerberg, 
CH-8093, Zurich, Switzerland, - (agruen, hanuscht)@geod.baug.ethz.ch 
SS-9: Cultural Heritage Recording and Silk Road 
KEY WORDS: Cultural Heritage, Modelling, Reconstruction, Visualization, Photo-realism, Texture, Close-Range, Non-Metric 
ABSTRACT: 
This paper reports about our efforts to generate a 3D computer model of the Small Buddha of Bamiyan, Afghanistan. This was one 
of two famous standing Buddhas, both of them being destroyed by the Taleban in March 2001. This adds to our previous work on 
the reconstruction of the Great Buddha and the precise modelling of the larger landscape around it. The main problem here was the 
lack of one suitable dataset (e.g. old images) which could have been used alone for reconstruction. So we had to resort to different 
kind of data, in order to make good use of everything that was available: Two old amateur images, taken before the destruction, 
newly made images of the now empty niche (acquired for the 3D modelling of the niche) and a given contour map, produced from 
the metric images of a past photogrammetric campaign of the University of Kyoto. We describe the fairly complex procedures that 
enabled us to get, despite the insufficient material, quite acceptable results. This shows again how advanced photogrammetric 
techniques can be applied to reconstruct lost objects. 
1. INTRODUCTION 
The valley of Bamiyan, Afghanistan, situated in the middle of 
the Silk Road, was one of the major Buddhist centers from the 
second century up to the time when Islam entered the area in 
the ninth century. Two big standing Buddhas were carved out 
of the rock, probably during the fourth and fifth centuries AD. 
The larger statue was 53 meters high while the smaller Buddha 
measured 38 m. The two colossal statues were demolished in 
March 2001 by the Taleban, using mortars, dynamite, anti 
aircraft weapons and rockets. At that time we were looking for 
a suitable dataset to demonstrate our procedures of 
reconstructing lost objects from old images. We picked the 
Bamiyan case because it was both technically challenging and 
practically relevant. 
In former publications we described the successful 3D computer 
reconstruction of the Great Buddha, together with the now 
demolished frescos, the now empty niches, the complete rock 
face and a large terrain area around it, using different kinds of 
images, at different resolutions and quality (Gruen et al., 2003, 
2004, 2006, Remondino and Niederoest, 2004). The results that 
we present here has to be seen as an extension of our previous 
work to reconstruct the archaeological site of Bamiyan. This 
paper describes the reconstruction of the Small Buddha statue 
and its integration into the current empty niche model. 
Two image datasets and an existing contour drawing were used. 
First, a series of nine terrestrial images, acquired in August 
2003, using a digital Sony Cybershot camera with 6 Megapixel 
image format was used to cover the empty niche after the 
destruction of the statue. The second dataset consists of two 
pictures, acquired using a 5.5 x 5.5 cm 2 analogue YASHICA 
frame camera with a focal length of around 80 mm in 1975 by a 
tourist, covering the statue before demolition. This dataset is 
characterised by very oblique viewing directions and a small 
base-distance ratio. In addition, the camera parameters, except a 
rough approximation of the focal length, were unknown. 
The orientation of the niche dataset, including estimation of the 
camera parameters was done in a previous work. Ground 
control points, acquired during the field campaign, using a total 
station, were used to define the scale and the common 
coordinate system for consistent representations of all datasets 
in the final model. In addition, a surface model of the empty 
niche was generated in the past. 
The orientation of the second dataset was done using different 
software systems. Special attention was paid to the calculation 
of the camera parameters. To define the scale of the model, 
control points, derived from the niche dataset, were introduced 
into the orientation procedure. 
Afterwards an automatic surface model generation was 
conducted using our own software SAT-PP (SATellite image 
Precision Processing). Because of the weak acquisition 
geometry, only the breast and leg parts of the statue could be 
reconstructed reliably. The more complex structures, especially 
the regions around the arms and the folds of the clothing, 
required manual measurements and editing. Shoulders, face and 
feet are not well enough visible in the images. To fill these 
regions, an existing contour map, originally produced by an 
earlier Japanese photogrammetric mission to Bamiyan, was 
used. After a rough georeferencing, the contour lines of the 
problematic regions as well as the breast and the upper leg part 
were digitized. A first estimation of the equidistance was 
conducted using the generated body model. Therefore, a second 
surface model of the digitized regions could be generated. 
The two surface models, one from the image measurements and 
one generated using contour lines, were co-registered using 
Geomagic Studio. Since the two datasets were available in the 
same coordinate system, a selective fusion could be conducted 
to generate a consistent depth model of the statue. 
T'm 
«. 'sub*; 
1