MARKOV RANDOM FIELDS (MRF)-BASED TEXTURE SEGMENTATION
FOR ROAD DETECTION
Ralf Reulke a , Artur Lippok b
‘‘Humboldt-University Berlin, Department of Computer Science, Computer Vision
Unter den Linden 6, 10099 Berlin, Germany - reulke@informatik.hu-berlin.de
b Hilbertstr. 22a, 12307 Berlin, Germany - lippok@gmx.de
Commission III, WG III/5
KEY WORDS: Mapping, Photogrammetry, Vision, Fusion, Identification, Monitoring, Real-time, Platforms
ABSTRACT:
Traffic observation from airplane platforms using digital cameras is a fairly new application of Video Image Detection Systems
(VIDS). These systems are also particularly interesting for observations at major events or in catastrophe situations. Applied systems
are small or medium-sized panchromatic cameras. In order to concentrate on road image processing algorithms were applied to
distinguish between urban and surrounding vegetation areas. Therefore, texture-based segmentation seems to be an adequate
approach, because of the panchromatic images. Attention is focused on MRF, which are examined concerning their use for texture-
based extraction of street space in aerial photographs.
KURZFASSUNG:
Der Einsatz von digitalen Kameras für Verkehrsbeobachtungen von Flugzeugplattformen ist eine relativ neue Anwendung von
Videodetektionssystemen (VIDS). Diese Systeme sind auch besonders interessant für Überwachungsaufgaben bei
Großveranstaltungen oder in Katastrophensituationen. Die angewendeten Systeme sind kleine oder mittelgroße panchromatische
Kameras. Bildverarbeitungsalgorithmen zur Unterscheidung von urbanen und umgebenen Vegetationsgebieten wurden angewendet,
um die Extraktion von Fahrzeugen lediglich auf den Straßenbereich zu fokussieren. Aufgrund der panchromatischen Bilder erscheint
die texturbasierte Segmentierung dafür der adäquate Ansatz. Besonderes Augenmerk gilt den MRF, die hinsichtlich ihrer Nutzung
für texturbasierte Extraktion von Straßenraum in Luftbildern untersucht werden.
1. INTRODUCTION
Traffic observation from airplane platforms using digital
cameras is a new application of Video Image Detection Systems
(VIDS) as already used for regular traffic observations.
Monitoring from airplane platforms is also particularly
interesting for observations at major events or in catastrophe
situations. An overview of current airborne and spacebome
monitoring systems and related activities can be found in (Hinz
et al., 2006).
The investigations described here, are based on the ANTAR-
system and the evaluation software “Traffic Finder” developed
by the German Aerospace Center. The system provides real
time extraction of traffic data. ANTAR includes a digital mega
pixel camera and an inertial measurement unit. The acquired
images are transmitted to the ground and analysed by the
software to extract cars and their features (Ruhe et al, 2007).
One essential feature of such a system is the immediate
knowledge of extracted cars or derived traffic parameters of
street sections, which can be annotated in digital street maps
(e.g. made by NAVTEQ). To improve the effectiveness of
image processing algorithms, irrelevant traffic areas are omitted.
In the presented example the observation area is defined with
help of digital street maps in combination with a digital
elevation model. For that purpose the camera orientation is
exactly determined. Unfortunately, the precision of roadway
geometry is restricted because of navigation errors, limited map
accuracy or map availability. Therefore, the area with occurring
traffic shall be exactly determined using additional image
processing algorithms. Utilized camera systems are small or
medium-sized panchromatic cameras. Image processing is
restricted to texture-based segmentation approaches due to the
panchromatic images.. Thus multispectral and colour
approaches are not applicable.
Segmentation is a commonly used term for identifying
differences between particularly interesting and uninteresting
objects as well as distinguishing foreground from background
content. In order to identify possibilities for acquisition of scene
information by digital images an analysis of the principle
features of these images is required. In this regard, textures are
the only possibility to derive information from imagery, besides
the grey or colour values and structural features. For this reason,
texture analysis methods were used since the beginning of
digital image processing. There are continuous suggestions for
new algorithmic approaches to texture-based scene analyses.
Thereby, also Laws-Energy, Fuzzy- and Fourier approaches are
analyzed besides first and higher order statistics.
Road detection is a prerequisite for traffic observation from
airplane platforms. The paper discusses different approaches to
texture-based feature extraction and segmentation. Standard
algorithms are examined first. Caused by the complicated
observation conditions these methods cannot be used
successfully.
615
