In: Stilla U, Rottensteiner F, Paparoditis N (Eds) CMRT09. IAPRS, Vol. XXXVIII, Part 3/W4 — Paris, France, 3-4 September, 2009 
103 
CLASSIFICATION SYSTEM OF GIS-OBJECTS USING 
MULTI-SENSORIAL IMAGERY FOR NEAR-REALTIME DISASTER MANAGEMENT 
Daniel Frey and Matthias Butenuth 
Remote Sensing Technology 
Technische Universität München 
Arcisstr. 21, 80333 München, Germany 
daniel.frey@bv.tum.de, matthias.butenuth@bv.tum.de 
KEY WORDS: System, Classification, Statistics, Multisensor, Integration, GIS, Disaster 
ABSTRACT: 
In this paper, a near-realtime system for classification of GIS-objects is presented using multi-sensorial imagery. The system provides 
a framework for the integration of different kinds of imagery as well as any available data sources and spatial knowledge, which 
contributes information for the classification. The goal of the system is the assessment of infrastructure GIS-objects concerning their 
functionality. It enables the classification of infrastructure into different states as destroyed or intact after disasters such as floodings 
or earthquakes. The automatic approach generates an up-to-date map in order to support first aid in crisis scenarios. Probabilities are 
derived from the different input data using methods such as multispectral classification and fuzzy membership functions. The main core 
of the system is the combination of the probabilities to classify the individual GIS-object. The system can be run in a fully automatic 
or semi-automatic mode, where a human operator can edit intermediate results to ensure the required quality of the final results. In this 
paper, the performance of the system is demonstrated assessing road objects concerning their trafficability after flooding. By means of 
two test scenarios the efficiency and reliability of the system is shown. Concluding remarks are given at the end to point out further 
investigations. 
1 INTRODUCTION 
A significant increase of natural disasters such as floodings and 
earthquakes has been observed over the past decades (Kundzewicz 
et al., 2005). There is no doubt that the disasters’ impact on the 
population has dramatically increased due to the growth of pop 
ulation and material assets. The regrettable death of people is 
accompanied by heavy economic damage, which leads to a long 
term backslide of the regions hit by the disaster. This situation 
calls for the development of integrated strategies for prepared 
ness and prevention of hazards, fast reaction in case of disasters, 
as well as damage documentation, planning and rebuilding of in 
frastructure after disasters. It is widely accepted in the scientific 
community that remote sensing can contribute significantly to all 
these components in different ways, in particular, due to the large 
coverage of remotely sensed imagery and its global availability. 
However, time is the overall dominating factor once a disaster 
hits a particular region to support the fast reaction. This becomes 
manifest in several aspects: firstly, available satellites have to be 
selected and commanded immediately. Secondly, the acquired 
raw data has to be processed with specific signal processing algo 
rithms to generate images suitable for interpretation, particularly 
for Synthetic Aperture Radar (SAR) images. Thirdly, the inter 
pretation of multi-sensorial images, extraction of geometrically 
precise and semantically correct information as well as the pro 
duction of (digital) maps need to be conducted in shortest time- 
frames to support crises management groups. While the first two 
aspects are strongly related to the optimization of communication 
processes and hardware capabilities, at least to a large extend, fur 
ther research is needed concerning the third aspect: the fast, inte 
grated, and geometrically and semantically correct interpretation 
of multi-sensorial images. 
Remote sensing data was already used in order to monitor natural 
disasters in the year 1969 (Milfred et al., 1969). Particularly, in 
the case of flooding a lot of studies are carried out to infer in 
formation as flood masks from remote sensing data (Sanyal and 
Lu, 2004). The flooded areas can be derived from optical im 
ages (Van der Sande et al., 2003) as well as from radar images 
(Martinis et al., 2009) via classification approaches. Zwenzner 
(Zwenzner and Vogt, 2008) estimates further flood parameter as 
water depth using flood masks and a very high resolution digital 
elevation model. Combining this results with GIS data leads to 
an additional benefit of information and simplifies the decision 
making (Brivio et al., 2002, Townsend and Walsh, 1998). The 
combination of the GIS and remote sensing data is often carried 
out by overlaying the different data sources. But, there are only 
few approaches which use the raster data from imagery to assess 
the given GIS data. In (Gerke et al., 2004, Gerke and Heipke, 
2008) an approach for automatic quality assessment of existing 
geospatial linear objects is presented. The objects are assessed 
using automatically extracted roads from the images (Wiedemann 
and Ebner, 2000, Hinz and Wiedemann, 2004). However, in case 
of natural disasters the original roads are destroyed or occluded 
and, therefore, it is not possible to extract them using the original 
methods. Hence, new approaches have to be developed which 
assesses damaged and occluded objects, too. The integration and 
exploitation of different data sources, e.g. vector and image data, 
was discussed in several other contributions (Baltsavias, 2004, 
Butenuth et al., 2007). However, there is a lack of methods which 
assess the GIS data concerning its functionality using imagery 
(Morain and Kraft, 2003). 
In this paper, a classification system using remote sensing data 
and additionally available information is developed to assess GIS- 
objects. The main goal of the system is the automatic classifica 
tion and evaluation of infrastructure objects, for example the traf 
ficability of the road network after natural disasters. However, 
the presented system can be transferred to other scenarios, such 
as changes in vegetation, because its design is modular. A focus 
is the integrated utilization of any available information, which 
is important to ease and speed up the classification process with 
the aim to derive complete and reliable results (Reinartz et al., 
2003, Frey and Butenuth, 2009). In comparison to the manual 
interpretation of images the presented systems is very efficient,