XXIX-B3, 2012 
natic Line Matching 
p. 666—671. 
construction methods 
1, Photogrammetric 
Vo: 7, 705-714. 
'ercity modeller. In:1 
"mote Sensing 
h 2006. 
D building model 
ground plans. In: 
and Remote Sensing, 
    
  
GESTALT GROUPING ON FACADE TEXTURES FROM IR IMAGE SEQUENCES: 
COMPARING DIFFERENT PRODUCTION SYSTEMS 
E. Michaelsen !, D. Iwaszczuk., L. Hoegner”, BB. Sirmacek?, U. Stilla? 
! Fraunhofer-IOSB, Gutleuthausstrasse 1, 76275 Ettlingen, Germany, eckart.michaelsen(g)iosb.fraunhofer.de 
? Technische Universitaet Muenchen (TUM), Photogrammetry and Remote Sensing, 80333 Muenchen, Germany 
? German Aerospace Center (DLR), Remote Sensing Technology Institute, 82234 Wessling, Germany 
Commission III, WG III/4 
KEY WORDS: Facade recognition, thermal imagery, production systems 
ABSTRACT: 
The façades of buildings are almost always organized according to Gestalt principles such as good continuation, repetition in 
similarity, or symmetry etc. Coding such principles in production systems yields a very flexible frame to explore the usefulness of 
such principles in automatic facade understanding. Capturing images and image sequences of façades in the thermal domain and 
understanding such data is of importance e.g. for energy saving. In this contribution two different production systems are compared 
using the same data and interpreter. 
1. INTRODUCTION 
1.1 Thermal Textures 
Thermal textures on facades of buildings are of growing 
interest. In thermal infrared (IR) images damages and weak 
spots in building hull (especially in building insulation) and 
heat waste can be observed. Thanks to combination with a 3D 
building model the spatial reference of the IR images is given. 
In urban areas with narrow streets it is often not possible to 
capture a whole fagade in one frame. Then the picture can be 
stitched from a video. Figure 1 shows such image which 
actually results from texturing a 3D building model by 
projecting suitable image information from a vehicle mounted 
thermal video camera (Hoegner & Stilla U, 2007). 
  
Figure 1. A thermal façade texture 
Of course artefacts from stitching cannot be completely 
avoided. But, obviously in such textures heat leakages can be 
detected and the heat bridges can be stored together with 3D 
building data. Our goal is to proceed in automation of this 
process. 
There are two reasons why windows should be detected in IR 
textures and excluded for the inspection. First, in thermal 
Imagery glass reflects the surrounding, e.g. sky, a neighbouring 
building and trees, and shows false results for the temperature 
measurements. Second, windows can influence automatic heat 
leakage detection (Hoegner & Stilla U, 2009) and lead to false 
results. Therefore a method for window detection in IR images 
is needed. 
1.2 Gestalt Grouping 
Façades are man-made objects and display strong gestalt 
structure, such as ordering according to lattice or symmetry 
principles. This also holds for the thermal spectral domain. The 
laws of Gestalt grouping are known for about hundred years, 
namely “good continuation”, “repetition in similarity", 
“symmetry — mirror or rotational” and proximity (Wertheimer, 
1923). Automatic Gestalt grouping can e.g. be performed by 
tensor voting or accumulator methods. The state-of-the-art has 
e.g. been presented at the symmetry competition along with the 
CVPR (Liu & Rauschert, 2011). 
1.3 Related Work 
In the last decade facades classification has drawn particular 
attention. Pu & Vosselman (2009) fused laser data and close- 
range images to reconstruct building facade details. They 
extracted windows and doors in both close-range optical and 
laser images by using Hough accumulation of lines. Detected 
windows and doors helped them to register close-range optical 
and laser images. That shows the importance of the facade 
classification study in three-dimensional city modeling. 
Burochin et al. (2009) proposed a segmentation method to 
detect repetitive structures like windows in close-range optical 
images. For segmentation they defined a model by considering 
shape and reflectance of a window. Then they applied matching 
process to find correspondence between model and image. In 
(Ali et al., 2007), a summary of the researches on window 
detection has been given. They also proposed a window 
detection system based on cascade classifiers. In a following 
study, Ali et al. proposed a system to detect windows in laser 
scanner data. They use depth variations to detect windows (Ali 
et al., 2008). Lee & Nevatia (2004) proposed a robust system to 
detect windows in optical images. They extracted window