Paul Pope
INITIAL TESTING OF A METHOD TO GEOGRAPHICALLY REGISTER AIRBORNE SCANNER
IMAGERY THROUGH PARAMETRIC MODELING WITH IMAGE-TO-IMAGE MATCHING
Paul A. Pope, Frank L. Scarpace
Environmental Remote Sensing Center
University of Wisconsin
Madison, Wisconsin, U.S.A.
papope @ students. wisc.edu
scarpace @macc.wisc.edu
KEYWORDS: whiskbroom, georeference, orthophoto, DEM, numerical simulation
ABSTRACT
This paper presents the results from initial testing of a new method for georeferencing airborne multispectral scanner
imagery. The method is called "parametric modeling with image-to-image matching" (PMIIM). The method takes a
error prone estimate of the aerial platform’s trajectory as input, and outputs a corrected trajectory. The corrected
trajectory can then be used to georeference the airborne imagery through another ray-tracing code. The method was
tested by synthesizing an airborne scanner image through the use of mathematically generated trajectory functions, A
discussion of how spatial autocorrelation affects the method is discussed. Pre-processing steps and justification for the
choice of input parameters is explained. The method was able to decrease the planimetric error from 34 meters to ?]
meters. Avoiding the coupling between some of the exterior orientation parameters enabled the run time to be reduced
dramatically, but incurred a decrease in planimetric accuracy.
1. BACKGROUND
This research concerns itself with addressing the problem of georeferencing multispectral imagery acquired primarily
by a whiskbroom scanner with a large field-of-view carried aboard an airborne platform flying at a relatively low
altitude. A previous paper has provided more details concerning the various methods described in the literature for
addressing this problem, as well as the justification, objectives, design, and implementation of a new solution (Pope and
Scarpace, 2000). Only a brief outline of this new philosophy is provided herein. This paper describes proof-of-concept
testing of this new method through numerical modeling and accuracy assessment of the results.
Compared to frame-based (e.g. aerial photography) and line-based (pushbroom) imaging systems, the problem of
georeferencing imagery acquired by a point-based (whiskbroom) scanner is the most difficult, due to the highly
dynamic nature of the viewing geometry. Large planimetric distortions are incurred through the inter-play between the
scanner dynamics, platform instability, and the terrain. Research to address this problem has given rise to two largely
disparate approaches; parametric and non-parametric.
The first is the pursuit of parametric (geometric, deterministic) methods. Examples of these methods are
photogrammetric (McGlone and Mikhail, 1985) and "direct georeferencing" (Schwartz, et al., 1993; Meyer, 1994)
These methods may also be categorized as "hardware" solutions, since they rely on highly accurate and temporally fine
measurements of the platform's trajectory (e.g. INS/GPS), or at least a trajectory "seed" which is iteratively improved
through a least squares bundle adjustment. Also, parametric methods often require a digital elevation model (DEM) to
obtain the highest planimetric accuracy.
The second is the pursuit of non-parametric (image-warping, empirical) methods. Examples of these methods are the
piece-wise application of bi-variate mapping functions (Ji and Jensen, 2000), iterative Delauney triangulation (Chen and
Rau, 1993), and multi-quadratic rectification (McGwire, 1998). These methods may also be categorized as "software
solutions, since they do not rely on any hardware-based measurements of the platform's trajectory. However, they often
depend on a reference image or a map to aid in defining ground control points (GCPs). The manual definition of GC
is known to be labor intensive. Automated image-to-image matching can aid in defining control.
A new solution has been developed which takes a "hybrid" approach and lies between these two extremes. It is
fundamentally parametric (ray-tracing) in nature, but is autonomous in defining control and takes as inputs the a prion
information typically used by both non-parametric and parametric methods. The proposed solution is to use a reference
image (e.g. a digital orthophoto or satellite image) and a DEM to define a "virtual landscape" over which à
mathematical model of the scanner and its airborne platform are "flown". A trajectory "seed" provides an initial
732 International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B3. Amsterdam 2000.
estil
dire
estii
con:
An
synt
seec
pert
usec
met]
ima;
2.0
The
airb
mod
the
info
exte.
Nois
hard
metl
ray-
loca
perti
ES
3.1 (
The
MH;
with
Thre
(Res
"cre:
synt
scan
pixel
traj e
Corre
scan
The
of hi
"drif
land