AUTOMATIC MEASUREMENT OF ROAD WIDTHS IN COLOUR STEREO SEQUENCES ACQUIRED BY A MOBILE
MAPPING SYSTEM
Krzysztof Gajdamowicz
Inst, of Geodesy and Photogrammetry
Royal Institute of Technology
Sweden
krzygajd@geomatics.kth.se
Commission II, Working Group 1
KEY WORDS: road inventory, automatic colour image analysis, colour based segmentation
ABSTRACT
In 1995, the Department of Geodesy and Photogrammetry together with the Swedish National Road Administration (SNRA) and the
Swedish National Board for Industrial and Technical Development (in Swedish: NUTEK) set up a project for the development of
methods for automatic analysis of the data acquired by Mobile Mapping Systems (MMS). Such methods are planned to be used for fast
and cost effective data collection for the National Road Data Base (NRDB).
Colour stereo sequences recorded by MMS are georeferenced and analysed in a post-process to obtain geometric and semantic
information about roads and objects in their surrounding. This paper presents methods for automatic measurement of road widths from
colour stereo georeferenced images. It also presents an evaluation of the automatic method for road width measurement in comparison
with semi-automated. Additionally, this paper discusses the advantages and limitations of the method. Finally, recommendations and
improvements of the method are presented.
The automatic measurement of road widths is based on step-wise processing of colour georeferenced stereo sequences. It employs colour
based segmentation for localisation of the road, image filtering for extraction of road boundaries, image matching for 3D positioning of
road boundaries and computation of 3D vector between the road boundaries to determine the width of the road. The final results are
revised and checked for gross errors. If any outliers are found, the measurements are repeated by semi-automatic stereo measurement,
where for a manually selected point in one stereo image the corresponding point is found by feature based matching in colour space.
After the matching procedure is completed, the road width is computed in the global co-ordinate frame.
1 INTRODUCTION
Inventory of roads and maintenance of the road data is a primary
task for the Swedish National Road Administration (SNRA).
Information about the road and road environment needs to be
acquired and frequently updated. Main attributes in the Road
Data Bank (RDB) are the nodal points, i.e. road junctions, and
the distance between them. The position information and the
attributes of the inventory objects like kerbs, manholes, road
signs, road widths, etc. are georeferenced using distance from
road junctions.
Presently, measurements of the road widths are performed in a
traditional way using survey tape. Such measurements are time
consuming and often require interference in a traffic flow. In case
of very trafficked roads measurement of the road widths requires
an alternative method.
One of then! is aerial photogrammetry. Up to now several
algorithms for automatic road extraction from aerial photographs
were presented, e.g. Mayer et al. (1998). Unfortunately, these
methods still require more research and optimisation in order to
be robust. Moreover, aerial photogrammetry is an expensive
method that has low repeatability, therefore it is not well suited
for the purpose of road width measurement.
The other way to acquire road widths is to use Mobile Mapping
System (MMS) and digital photogrammetry. MMS like On-Sight
developed by Transmap Corp. is a system, which integrates GPS,
INS, odometer and colour digital cameras on a van. Such a
system allows acquiring stereo georeferenced images of the
roads while driving with speed up to 80 km/h. Road widths are
measured in a post-process using georeferenced images and
stereo photogrammetrical methods. Monotonous image point
measurement procedure requires automation. Habib (1994)
described in his thesis an algorithm for automatic extraction and
matching of the road edges (painted lines) using MMS
georeferenced images and data association technique. Tao (1996)
presented an integrated approach to reconstruct centrelines using
stereo image sequences and MMS trajectory calculated from
GPS/INS data. Gajdamowicz (1998) presented a semi-automatic
approach for point and vector measurement and an automatic
approach for road sign inventory.
The primary goal of this paper is to present an approach to
measure road widths automatically. The secondary goal of this
7B-2-1