In: Paparoditis N., Pierrot-Deseilligny M.. Mallet C. Tournaire O. (Eds). IAPRS. Vol. XXXVIII. Part ЗА - Saint-Mandé, France. September 1-3. 2010 
QUALITY MEASURE FOR TEXTURES EXTRACTED FROM AIRBORNE IR IMAGE 
SEQUENCES 
D. Iwaszczuk, U. Stilla 
Photogrammetry and Remote Sensing, Technische Universitaet Muenchen (TUM) - (iwaszczuk, stilla)@bv.tum.de 
Commission III, WG III/5 
KEY WORDS: Infrared Images, Image Sequences, Automated Texture Mapping, 3D Building Models, Oblique View 
ABSTRACT: Energy and climate changes are big topics in near future. In the European countries a significant part of consumed 
energy is used for heating in the buildings. Much effort is required for reducing this energy loss. Inspection and monitoring of 
buildings contribute in further development saving energy. For detection of areas with the highest loss of heat thermal infrared (IR) 
imaging can be used. For such inspection a spatial correspondence between IR-images and existing 3D building models is helpful. 
This correspondence can be created by geo-referencing of the images. Then, the 3D model can be projected into the image and for 
each surface of the model a region of the image can be selected for texture. In this paper a concept for the texture mapping using IR 
image sequences is introduced. Emphasis is placed on analysis of texture quality and resolution changes in oblique view images. The 
influence of the angular aperture and inclination angle on the effective texture resolution is discussed. A formula for calculation of 
texture quality is proposed. For experiments with data an IR image sequence of test area “Technische Universitet Muenchen” in 
Munich is used. 
1. INTRODUCTION 
1.1 Motivation 
Due to the climate changes and increasing energy costs became 
the energy efficiency an important topic. In the European 
countries 40 % of the energy is consumed by buildings, whereas 
47% is used for heating. Much effort is required for reducing 
the energy loss. Inspection and monitoring of buildings 
contribute in further development saving energy. Thermal 
infrared (IR) imaging can be used to detect areas with highest 
loss of heat. Nowadays usually single IR images are analyzed, 
without reference to the geometry of the captured scene, often 
manually. It does not allow a combination of thermal data with 
other imagery or semantic information stored in a spatial data 
base, is time consuming and not applicable for large areas. For 
geo-referencing the IR images can be combined with three- 
dimensional geometry of the buildings as textures and in these 
textures heat leakages can be detected (Hoegner & Stilla, 2009). 
Such monitoring of whole cities allows creation of an open 
geographic information systems (GIS), where citizens could 
view the energy loss of their houses. 
High resolution, terrestrial IR images can be acquired using a 
mobile platform and used for texturing of façades (Hoegner & 
Stilla, 2007). However, these images do not capture the roofs 
and façades from inner yards or building rings. To complete the 
coverage of the buildings with textures oblique view images 
from an airborne platform can be used (Grenzdoerffer et al., 
2008; Stilla et al., 2009; Kolecki et al., 2010). 
1.2 Related work 
In recent years the process of automatic texture mapping has 
been frequently discussed within photogrammetry and computer 
vision. The major problems connected to the automation of the 
texturing are: 
• geo-referencing, 
• self-calibration of the camera, 
• automatic visibility checking and self occlusions, 
• elimination of extrinsic occlusions, 
• selection of the best image for the texture from an available 
set of images. 
For geo-referencing of aerial images, the exterior orientation 
(ExtOri) parameters of the camera are needed. Approximated, 
but stable position and orientation of the sensor can be directly 
determined using global positioning system (GPS). An inertial 
navigation system (INS) provides good short-term accuracy, but 
in a longer time a systematic drift occurs. Thus, the combination 
of GPS and INS allows to avoid the INS drift and to bridge any 
loss of satellite signal by GPS (Yastikli & Jacobsen, 2005). 
Unfortunately, often the camera position and orientation are not 
identical with position and orientation registered by GPS/INS 
device. In that case the estimation of the misalignment angles 
(boresight parameters) and the lever arm vector is necessary (E 
Yastikli & Jacobsen, 2005; Eugster & Nebiker, 2007; Stilla et 
al., 2009; Kolecki et al., 2010). Furthermore, most of building 
models are stored in national coordinates, while GPS/INS 
navigation uses geographic coordinate system. Skaloud & Legat 
(2008) present necessary transformations between both 
coordinate systems. 
However, the accuracy of direct geo-referencing is not 
appropriate for a precise texture mapping with high-resolution 
images. Additionally, the interior orientation parameters are 
usually only approximately known. Therefore a self-calibration 
algorithm is required. Frueh et al. (2004) proposes an approach 
based on matching of line segments with model edges. In this 
method the edges are extracted in the image and the model is 
projected into the image from a random pose of the camera. For 
this pose a rating based on line matching is calculated and this 
procedure is repeated. The pose with the highest rating is 
chosen for texture mapping. However, computational effort of 
this method is very high. Ding & Zakhor (2008) present an 
algorithm for self-calibration divided in two steps. In the first 
step vanishing points corresponding to vertical lines and GPS-