ation,
fying
earch
S for
tional
oma
)bject
-RTI
ampa
.ange
thods
TV.
detic
ensor
ping
yrmat
ange
MR
ingle
Close
1ortis
ation
ig of
sium
; and
1994.
deck
sium
; and
mple
ssion
sion,
urne:
erior
and
s of
125-
GEOMETRIC INFORMATION FROM SINGLE UNCALIBRATED IMAGES OF ROADS
L. Grammatikopoulos *, G.E. Karras *, E. Petsa "
* Department of Surveying, National Technical University of Athens, GR-15780 Athens, Greece (gkarras@central.ntua.gr)
? Department of Surveying, Technological Educational Institute of Athens, GR-12210 Athens, Greece (petsa@teiath.gr)
KEY WORDS: Projective, Geometry, Calibration, Video, Sequences
ABSTRACT
Video sequences of road and traffic scenes are now used for various purposes. The framework of this research on the metric potential
of single uncalibrated images is road mapping and studies of the traffic character of freeways. In the first case, an approach has been
developed to extract lane width in straight road segments exploiting sequences from a forward looking camera. Apart from an initial
reference width, necessary for calibrating camera height, no intrinsic or extrinsic calibration is required if frontal image acquisition is
assumed. This approach, making use of the vanishing point of the road, gave an accuracy better than 5 cm in lane width. The second
technique regards the measurement of vehicle speed, given the time interval between frames of a stationary camera tilted downwards.
Here, too, the vanishing point in the direction of the road is used, with the vanishing point of the orthogonal direction assumed at
infinity. Given one known ground distance along the road axis, the projective distortion of the ground plane is removed, allowing an
affine rectification and, thus, 1D measurement in the correctly scaled road direction. This approach, evaluated against rigorous 2D-2D
projective transformation and GPS measurements, has given a satisfactory estimated accuracy in vehicle speed of about 3 km/h.
1. INTRODUCTION
Video sequences of roads are increasingly used in various con-
texts. For instance, preserving road network operability is today
a basic consideration, mainly focusing on road safety. Road re-
cording and mapping systems are needed to provide the missing
data, chiefly as regards older roads constructed and maintained
under various jurisdictions. Indeed, several mobile mapping and
video-logging systems have been reported (Tao and El-Sheimy,
2000). The former integrate multiple sensors with one, or more,
cameras providing geo-referenced image sequences and are cap-
able for precise 3D measurement. Video-logging systems, on the
other hand, are typically based on a single camera. Hence no 3D
measurement is generally possible in these cases, unless certain
geometric constraints are adopted (Tao, 2001). Most important
among them is the ‘flat-earth’ model.
The potential of automatic single-image approaches, relying on
different geometric assumptions, is being currently investigated
in the computer vision literature for various purposes. Probably
the main task here is the development of driver-assistance tools
or even autonomous road following. In this context, algorithms
have been presented for automatic lane and obstacle detection
(Bertozzi & Broggi, 1997, Enkelmann, 2001), the estimation of
lane shape and a vehicle’s road position or ‘ego-motion’ (Stein
et al., 2000; Southall & Taylor, 2001). Assuming constant road
width, Guiducci (1998) has reported on an algorithm for the 3D
road reconstruction from its image boundaries.
Of course, besides such ‘on-board’ approaches, video sequences
from a stationary camera are also used, e.g. with the application
of image processing and vision algorithms to traffic scenes for
queue detection and vehicle classification or counting (Dailey et
al., 2000). In particular, ordinary video cameras present certain
advantages over other techniques for monitoring vehicle speed,
e.g. with object tracking (Chun & Li, 2000), which represents a
crucial parameter in studying the traffic character of freeways.
Most monoscopic methods cited above are characterised by the
employment of cameras with known interior orientation (‘intrin-
sic camera’) parameters. Several camera calibration methods are
available, while approaches have also been proposed which use
the road itself as the calibration structure (Guiducci, 2000). But,
wherever possible, the potential of uncalibrated cameras should
be exhausted within flexible, low-cost approaches. For obvious
reasons, the exterior information required should also be kept to
a minimum to bypass the need for either point correspondences
and targets, or additional sensors providing camera attitude. In
speed measurement, for instance, the state-of-the-art in camera
technology adopted by transportation agencies requires detailed
intrinsic and extrinsic calibration of expensive cameras, and few
efforts have been made to measure speed using sequences of un-
calibrated images (Dailey et al., 2000; Pumrin & Dailey, 2002).
In this contribution approaches are presented and evaluated for
measuring two types of quantities with single image techniques:
lane width; and vehicle speed which, with known time intervals
between frames, is obviously a problem of measuring distances.
The video cameras used here are uncalibrated, in the sense that
camera constant, principal point location and the image affinity
parameters are irrelevant (radial lens distortion is not taken into
account). Regarding exterior orientation, following ‘reasonable’
assumptions are made here: the ground in front of the camera is
planar; cameras are fixed to the ground or the vehicle; images
have negligible rotations about the vertical and camera axes (cf.
Tao, 2001). Considering exclusively straight road segments, the
only requirements for the 2D-2D image-to-road transformations
in both cases (measurement of lane width and vehicle speed) are
one known distance. It is currently under investigation how the
approaches may be extended to include curving road segments
or significant rotations about the vertical camera axis.
Numerical results presented in the existing literature as evalua-
tions of algorithms measuring lane width and vehicle speed are
indeed sparse. Thus, aim of this contribution was to present the
mathematical models as well as to assess their performance with
sufficient measurements. No automatic extraction technique has
been applied here. These can be found in the existing literature,
though even relatively straightforward tasks, such as lane track-
ing, may well prove unexpectedly difficult due to changing road
and weather conditions (Southall & Taylor, 2001).
2. MEASUREMENT OF LANE WIDTH
To effectively manage existing road networks as regards opera-
bility and safety, sufficient information about their design para-
meters and installed equipment is needed; this concerns prima-
rily older roads for which there is a considerable lack of data. A
cost-effective road surveying system, designed at the Technical
-21-
