ic orienta-
om digital
blishes the
ddress the
utomation
‘his proce-
mented at
about the
on the van
are math-
1e ground.
extracted
age space.
the roads
nd. Thus,
ientataion
mall scale
'erlapping
'chniques,
d feature
ccessfully
1 between
991, Ack-
the exte-
gistration
of images, and the mapping from orthoim-
ages, have not reached advanced levels of au-
tomation. Operator intervention is still nec-
essary in these tasks. This group of tasks
has to be solved by determining the corre-
spondence between an image and a model of
the object space; for example, for the exterior
orientation of an aerial photograph one has
to determine a match between ground con-
trol points and their conjugate locations in
the image. Finding such a match is a very
difficult problem, because it requires match-
ing the 3-D ground representation of control
points, which is in vector form, to the 2-D im-
age representation of the same points, which
is in raster form.
In photogrammetric practice, the correspon-
dence between image and object spaces is es-
tablished by identifying and measuring con-
trol points in images manually In many
cases, such as in satellite images or small scale
aerial photographs, it is very difficult to iden-
tify control points due to the low resolution
of the images. This paper presents a general
procedure for exterior orientation of images
by matching features in the image with those
on the ground. This approach is extremely ef-
ficient for orienting satellite images and small
scale aerial photos. Thus, it has a great po-
tential for real time mapping applications.
During the past three years a mobile map-
ping system has been developed and imple-
mented by the Center for Mapping at The
Ohio State University. It collects informa-
tion about the environment of highways from
a driving van. The GPS receiver mounted
on the vehicle (GPSVan) determines the road
alignment in a ground coordinate system [No-
vak 1991]. These road alignments provide a
model of the road in object space. Digital
images of the same area can be analyzed to
extract roads automatically. By comparing
roads on the ground and in the images, a cor-
respondence between linear features in object
and image spaces can be established auto-
matically. Consequently, the automatic exte-
rior orientation of digital images using linear
features as control can be solved [Tankovich
1991, Mikhail 1993].
The system implemented for solving the cor-
respondence problem between linear features
in images and on the ground consists of three
modules. These modules are the ground mod-
ule, the image module, and the matching
module. They are discussed in the follow-
ing sections. Practical results are presented
in section 5 of this paper.
2. THE GROUND MODULE
The GPSVan captures a huge number of dis-
crete 3-D ground coordinates representing the
centerline of a road. The goal of this ground
module is to segment road centerlines and de-
scribe them mathematically. The first step
is the detection of the critical (important)
points of the road alignment. Next, the co-
efficients of the B-splines are computed such
that the critical points serve as break points
of the piecewise polynomials.
2.1 Detection of Critical Points
The detection of critical points for B-spline
representation is a crucial step, because these
points are representative of the original data.
Redundant points should be deleted, and im-
portant features of the original data should
be maintained.
The method implemented by the authors is
based on the local properties of the curve.
It depends on calculating the angles between
two line segments P; ,P; and P;P;,,. All
points P; for which this angle is less than a
threshold are deleted. This method has the
advantage of keeping only a few points for a
straight line, and many points for lines of high
curvature [Ballard and Brown 1982].
2.2 Cubic B-splines Representation
Splines are named after the draftman's de-
vice for drawing fair curves between specific
175
