AUTOMATIC ANALYSIS OF HIGHWAY
FEATURES FROM DIGITAL STEREO-IMAGES
Guangping He, Kurt Novak
Department of Geodetic Science and Surveying, Center for Mapping
The Ohio State University
Commission III
ABSTRACT
A mobile mapping system was developed by the Center for Mapping of the Ohio State
University to capture digital stereo-image sequences from a moving van. Each stereo-
pair is oriented and located in a global coordinate system by GPS and an inertial system.
Therefore, any point which appears in an image-pair can be located in a global
coordinate system. Techniques were developed to automatically recognize objects in
each image pair. Specifically, we concentrated on finding traffic signs and road edges.
; The edges of a highway can be easily extracted, if they are marked by bright lines. In
this case an horizontal edge operator is applied, and the bright markers are
automatically detected and approximated by straight lines in each image. An intersection
transfers the road edges to object space. If a starting point is defined, road edge can be
automatically traced until is vanishes. We analyzed milemarkers along railroad tracks
and were able to automatically read the log mile number.
Keywords: Digital Systems, Feature Extraction, Image Analysis, 3-D
1. BACKGROUND
The creation of geographic information systems (GIS)
requires enormous amounts of digital information. To date
most land related data-bases still rely on existing line maps
which are manually digitized. In order to collect digital
data faster and more accurately the combination of new
mapping sensors is necessary. An integrated system can
produce digital maps on-line from a moving vehicle,
therefore, we talk about "Real Time Mapping".
At the Center for Mapping of the Ohio State University
a number of integrated, mobile mapping systems have been
designed, assembled, and demonstrated. The most
successful system to date is the so-called GPS-Van
(Bossler, et. al. 1991). Its development was initiated by the
US Federal Highway Administration, 38 state
transportation agencies, as well as private companies. In
principal, it consists of three components: an absolute
positioning system, a relative positioning device, and tools
for gathering attribute data for the GIS (figure 1).
Figure 1: The GPS-Van integrates a digital stereo-vision
system with absolute positioning sensors.
119
The absolute positioning sensor is a combined
GPS/inertial surveying unit. Using differential GPS
processing the road-alignment can be mapped with an
accuracy of 1-3 meters in a global coordinate frame. The
inertial system, which consists of two gyros and a wheel
counter, takes over when satellite-lock is lost. For relative
positioning a stereo-vision system was mounted on the
GPS-Van. It yields 3-dimensional coordinates relative to
the vehicle; they can be transformed into a global system
by information collected by the absolute positioning
sensors. Finally, an analog video-camera and a touchscreen
are applied to collect attributes . All data captured by the
GPS-Van is immediately stored in a relational data-base
that was enhanced by an image management and analysis
system.
The GPS-Van's stereo-vision system permits the user to
locate any object in the field of view of both cameras in a
global coordinate system. The theory behind the
transformations involved is discussed in another paper (He,
et. al., 1992).
Here we concentrate on the automatic analysis of the
digital road images. The extraction and identification of
objects in one of the most challenging topics of our
research activities. It is absolutely necessary for a practical
utilization of the stereo-images, which represent such a
large amount of data that a human operator would not be
able to extract all information in a reasonable time. The
major advantage of detecting features in road images is that
we can well predict what kind of objects will appear in the
images-pairs, we can even predict their locations.
So far we have been successful in extracting road-edges
and center-lines, in matching almost any point in the stereo-
pairs, and in reading text and numbers in the digital images.
The named tasks are explained in this paper. The following
chapter gives a short description of the hardware
components used in the GPS-Van. In this conclusions we
talk about ongoing developments related to the image-
analysis software, and its full integration in a GIS.
