! 2004
by
da
1S
ed
PE
Y.
ial
Cr,
HN
for
96.
9s.
ors
oto
04,
ror
PE
ny.
rne
The
gy.
and
ON
49.
by
sion
rior
and
and
SEMI-AUTOMATIC GEO-REFERENCING IMAGES
IN MOBILE MAPPING
R. Roncella, G. Forlani
DICATA, Dept. of Civil Engineering, Parma University, Parma, 43100, Italy — (rroncell, gianfranco.forlani)@unipr.it
Commission III, WG 1
KEY WORDS: : Georeferencing, Image Matching, Image Sequences, GIS, Mobile Mapping, GPS/INS, Surveying
ABSTRACT:
This paper presents a purely photogrammetric strategy to orient short image sequences in a MM van equipped with two GPS
receivers and a pair of synchronized cameras but currently still lacking an IMU. The motivation for this is twofold: bridge over
short GPS outages, so increasing the vehicle productivity; improving the consistency of image georeferencing between consecutive
image pairs, which in relative terms is poor due to the limited accuracy of the GPS-supported camera parameters determination.
Drawing on techniques developed for structure and motion reconstruction from image sequences, a general method has been tailored
to the specific conditions of the MM imaging geometry, trying to ensure reliability of the matches and stability of the solution.
Though currently not all constraints between synchronous image pair are yet enforced, the first results suggest that the technique
may be working satisfactory, ensuring that the error propagation is within the specifications for GPS outages of about 100 m..
1. INTRODUCTION AND MOTIVATIONS
Since the advent oi digital photogrammetry it is apparent that
photogrammetry and computer vision increasingly share goals
and methods, though retaining their roots, i.e. stressing
accuracy and metrology aspects the former, focussing more on
real-time applications, image understanding and artificial vision
the latter. The progress in digital imaging sensors opens new
fields of application but demands for improvement in
automation of the processing, to cope with the amount of data
and to keep production costs in check. In this context, the
automation of image orientation and, to some extent, image
restitution, is being addressed in two ways: by developing
automatic methods based on image coordinates or by measuring
directly the exterior orientation elements.
As a well known example of the former approach, which has
been growing steadily in the last years, we may list structure
and motion (S&M) reconstruction from multiple views;
techniques proposed under this umbrella have been applied to
several fields, from visualization, archeological and
architectural surveying, computer graphics and so on.
On the other side, integrated GPS/IMU systems are being more
and more used in aerial and terrestrial application. The
increasing demand on database population and updating is
pushing demand towards using so-called high productivity
surveying vehicles or mobile mapping vehicles. Through
sensors and system integration (GPS, digital cameras, INS, laser
scanners... ), these vans can collect georeferenced data (mainly
images and 3D point clouds) at relatively high speed.
Somehow following a midway path, we are trying to make both
approaches to cooperate, to improve the reliability of a mobile
mapping system and its productivity. Since about one year, we
are developing at our Department a mobile mapping system for
the acquisition and updating of road databases and road
maintenance. The system specification are of an absolute
accuracy of about 1 m in horizontal, 5m in elevation and 10 cm
in relative accuracy (on the measurement of distances, mainly
the road width). Due to limits in funding, we are upgrading and
improving the on-board sensors in steps, trying to get the best
out of the available instrumentation, also by developing
software for image georeferencing from the navigation data and
for road database population from the oriented images. One of
the key modules of the software for image orientation, which is
the subject of this paper, is being developed to bridge
photogrammetrically over small GPS outages and improve the
restitution of points in asynchronous image pairs.
1.1 A mobile mapping van without an IMU
Currently our vehicle is just equipped with a pair of Leica SR
530 GPS receivers, mounted on the roof about 3 m apart. The
two receivers provide position, pitch and yaw of the vehicle, so
the roll angle currently cannot be determined. From the test
carried out, the accuracy of the yaw angle is around 0.1 degrees
or better, under good GPS conditions (no reliable figure has yet
been verified for pitch).
Two B/W Basler AF101 digital cameras with 8 mm focal lenght
and resolution of 1300x1030 pixels are mounted on the front,
with optical axes parallel and slightly inclined downwards and a
base of about 1.70 m. The cameras, with a pixel size of 6.7
micrometers, can acquire up to 12 tps at full resolution and are
synchronized to the GPS through the exposure signal sent to the
input event port of the receiver. Typical mission parameters are
an operating speed between 20-30 km/h and a frame rate of 2
Hz, with the GPS receiver acquiring at 10 Hz. After system
calibration, under good GPS conditions and on level road
sections, absolute accuracies in the range 20-50 cm have been
verified at distances up to 15 m; relative accuracy in a
synchronous stereo pair is better, from 2 to 5 cm across track at
distances below 10 m.
Obviously, besides the errors arising from the unknown roll
angle of the vehicle, the main problem is currently the lack of
an inertial measurement unit, capable to make up for the loss of
lock of the GPS signal due to occlusions caused e.g. by trees
and buildings: this severely limits the productivity of the
vehicle at the current stage of development. Though we are not
arguing that we can dispense with a IMU to reach a truly
operational level (and this is obviously our long term goal),
there is still a sizeable set of roads where we can perform a
