PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING
October 2016
775
Evaluation of Geometric Elements of
Repeat Station Imaging and Registration
Douglas A. Stow, Lloyd C. Coulter, Christopher D. Lippitt, Garrick MacDonald, Richard McCreight, and Nicholas Zamora
Abstract
Repeat station imaging (RSI) is an approach to precise aerial
image registration and change detection. It involves planning
and conducting aerial imaging flights such that the same sen-
sor returns to the same imaging stations over time, replicating
view geometry, and the resultant images are geometrically
processed on a frame-by-frame basis. The objective of this pa-
per is to elucidate and model the geometric elements associ-
ated with RSI capture and processing that influence the accu-
racy of co-registration. When each type of offset or difference
in orientation is modeled separately, relative displacement of
features in RSI pairs are linear functions of the varying orien-
tation parameters and focal length, with the exception of an
inverse square relationship with altitude above ground level;
the combination of higher altitude and longer focal length re-
duces image displacements approximately in half. The aver-
age difference (bias) between modeled and measured parallax
in ground dimensions is 0.1 m. Isolating the effects of orien-
tation and offset factors is a useful first step towards devel-
oping a robust analytical model of the geometry of RSI pairs.
Introduction
Tracking and monitoring dynamic Earth surface objects or
phenomena through multitemporal remote sensing is of-
ten referred to as image-based change detection. The term
“change detection” is commonly used, even if the application
objective goes beyond simple detection, to identify the type
or quantify the nature of earth surface changes (Jensen, 2007).
When the objective is to reliably capture changes of Earth
surface features or moving objects having sizes close to the
ground sampling distance of multitemporal imagery used for
analysis, time-sequential images must be accurately co-regis-
tered (Dai and Khorram, 1998). When very high spatial reso-
lution imagery is utilized, pixel-level co-registration accuracy
is difficult to achieve and errors can lead to a large number
of false change detections. This is particularly true for scenes
containing abrupt topographic relief and/or tall built features
(Toutin 2004).
The magnitude of signal and noise sources for multitem-
poral image differencing, the most common image-based,
pixel-level change detection approach, can be quantified as
(Stow 1999):
∆
∆
∆
∆
∆
∆
B
t
B
x
D
B
y
D R S
x
y
= −
−
+ +
(1)
where:
B
is the image brightness (i.e., digital number, radi-
ance, reflectance, surface temperature) value;
∆
∆
B
t
is the
discrete temporal change in image brightness (
B
) over time
interval
∆
t
, i.e., image difference;
∆
∆
B
x
and
∆
∆
B
y
are the
discrete spatial gradients of image brightness in the
x
(east-
west) and
y
(north-south) directions, respectively;
D
x
and
D
y
are misregistration magnitudes in
x
and
y
directions, respec-
tively;
R
is the temporal difference in
B
due to radiometric in-
consistencies (e.g., sensor noise, sensor calibration drift, and
differences in illumination and atmospheric optical proper-
ties); and
S
is the temporal change in
B
resulting from actual
land surface changes, i.e., the signal. Typically, the greatest
noise source is the magnitude of the product of local misregis-
tration error times the spatial gradient of image brightness in
the
x
and
y
coordinate directions. This means that pixel-level
change detection is most challenging for high spatial resolu-
tion imaging of scenes with heterogeneous landscape charac-
teristics. When co-registration errors are greater than one
or two pixels, the magnitude of
∆
∆
B
t
may be predominantly
influenced by the misregistration effect, resulting in falsely
detected changes and obscuring actual changes of interest.
Thus, a key to accurate and reliable change detection based
on high spatial resolution images is accurate co-registration of
multitemporal images.
Technological developments in digital camera sensors,
small manned and unmanned aircraft, global navigation
satellite systems (
GNSS
), and automatic tie point selection
and matching make it possible to capture and process hy-
per-spatial resolution, multitemporal aerial image data sets
in a relatively inexpensive and logistically feasible manner
(Pajares
et al.
, 2015; Zhang
et al.
, 2015). If pixel-level co-reg-
istration of image data supplied by these sensor systems can
be routinely achieved, the potential exists for conducting very
detailed aerial image-based change detection in support of a
variety of applications. Such potential applications include
but not limited to: surveillance in support of law enforcement
and national defense, search and rescue, monitoring of linear
utility and infrastructure assets, environmental monitoring,
and post-hazard damage assessment.
Almost exclusively in practice, airborne multitemporal im-
age sets are implicitly co-registered by subjecting each image or
set of images captured at different times to an orthorectification
Douglas A. Stow, Lloyd C. Coulter, and Garrick MacDonald
are with the Department of Geography, San Diego State Uni-
versity, 5500 Campanile Drive, San Diego CA 92182 (stow@
mail.sdsu.edu).
Christopher D. Lippitt is with the Department of Geography
University of New Mexico, Bandelier Hall West, Room 203,
Albuquerque, NM 87131.
Richard McCreight is with NEOS LTD., 2930 Horizon Hills
Drive, Prescott AZ 86305.
Nicholas Zamora is with Leidos, 10260 Campus Point Drive,
M/S C5, San Diego, CA 92121.
Photogrammetric Engineering & Remote Sensing
Vol. 82, No. 10, October 2016, pp. 775–685.
0099-1112/16/775–685
© 2016 American Society for Photogrammetry
and Remote Sensing
doi: 10.14358/PERS.82.10.677
