omplied by
pp.47-56.
. Structural
1) and GIS
ive, Lecture
| Information
'roceeding of
COSIT'95,
Frank, A.U.
AN INTEGRATED APPROACH TO ROAD CENTERLINE RECONSTRUCTION USING
STEREO IMAGE SEQUENCES FROM A MOBILE MAPPING SYSTEM
Chuang Tao
The University of Calgary, Department of Geomatics Engineering
2500 University Drive, NW, Calgary, AB, Canada T2N 1N4,
IWG V/III
KEY WORDS: Vision, automation, integration, modeling, object reconstruction, image sequence analysis, image matching.
ABSTRACT:
The reconstruction of 3D road centerlines becomes a physical problem of solving an energy-minimizing 3D B-splines shape model
based on "shape from sequences". The reconstruction is described as a process whereby a 3D road centerline shape model is
deformed gradually, driven by forces arising from object space (internal energy) and image space (external energy). The integration
of multiple constraints from a mobile mapping system is implemented. Recent test results demonstrate that this approach functions
reliably even in situations where the road condition is far from ideal.
1. INTRODUCTION
The management of vehicles and infrastructure requires a high
quality and up-to-date highway related spatial information
system. Road centerline information is very important for
generating road network information systems. It can be used to
compute road inspection parameters such as the longitudinal
profile and the surface deformation. The acquisition of up-to-
date road centerline data by conventional field survey is
prohibitive in cost and in actual environment reasons. Since
1992, The University of Calgary jointly with GEOFIT Inc. has
been developing a mobile mapping system, VISATTM, for fast
spatial information collection, especially for road inventory
(Schwarz et al., 1993; Li et al, 1994). In the system. CCD
digital cameras, mounted on the top of the van, have been
employed to collect stereo and sequential images of road
centerlines. The integration of the GPS and the INS has been
applied in vehicle location and image sequences
georeferencing. The information of road centerlines can be
extracted from images during the post-mission processing. The
first system demonstration was made in July 1993. The
prototype system was tested between October 1993 - October
1995. The production system is available since 1995.
The extraction of road centerline information from images has
been researched in vision-based vehicle navigation (Thorpe et
al, 1988; Schneiderman and Nashman, 1994). But the
automated reconstruction of 3D road centerlines from mobile
mapping systems to generate a road network information
system 1s still a new project (He and Novak, 1992). In our
system, not only 1s the centerline information required to be
extracted in an automatic way, but also the accuracy and
robustness of the extracted data should meet the requirements
of mapping applications. This greatly differs from the previous
research.
An integrated approach to road centerline reconstruction from
image sequences is proposed. In this approach, the
reconstruction of road centerlines from image sequences has
been considered as a problem on “shape from sequences.” The
problem is to synthesize road centerline information available
from successive images into a 3D shape model. Firstly, a 3D
physically-based shape model of road centerlines is set up. In
order to synthesize the constraint information coming from
object assumptions and image sequences into the model, the
857
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B3. Vienna 1996
model is defined as an active and deformable 3D curve. The
physically-based deformation mechanism has been endowed
with the model such that the model can be progressively
deformed under the action of internal and external constraint
forces. The extracted data of road centerlines from image
sequences can act as external energy which enforces the model
to deform towards its desired position. Internal energy arises
from smoothness constraints representing the natural
characteristic of the shape of road centerlines. It maintains the
a priori assumptions about the shape of the model. Under a
combination of the actions of internal and external forces, the
model will be deformed incrementally towards the final state at
which forces from different sources are balanced. The model
resulting at the end of an input sequence represents a 3D road
centerline shape.
This new approach leads a solution of the integration of
multiple constraints in the mobile mapping system: ego-motion
constraints (GPS/INS based vehicle trajectory determination
and image sequence georeferencing). object space constraints
(road model based reconstruction and model driven feature
extraction/matching), and image space constraints (stereo and
motion image geometry). To implement the approach, three
key problems are involved:
e How to set up an approximate 3D shape model of road
centerlines?
e How to synthesize multiple constraints into a shape
model?
e How to obtain the reliable constraint information (external
energy) from image sequences?
The above problems are addressed in section 2, 3 and 4,
respectively. The test results of real image sequences and
computational aspects of the approach are evaluated in section
5. Concluding remarks and future work are given in section 6.
2. VEHICLE TRAJECTORY DETERMINATION
FOR MODEL INITIALIZATION
The approximate 3D shape model of road centerlines forms the
basis of the implementation of “shape from sequences.” A
novel method for obtaining such an approximate shape model
is proposed. The kinematic vehicle trajectory is utilized to
generate an approximate shape model. In VISAT™, a three-