Remote Sensing Image Matching for Unpiloted
Aircraft based on SIFT Arithmetic Operators
Pei Liang/Liaoning Technical University
School of Geomatics
Fuxin, China
lgdpl@sina.com
Qi Yuanchen/Liaoning Technical University
School of Geomatics
Fuxin, China
qiyuanchen649@163.com
Zhao Hongying/Liaoning Technical University
School of Geomatics
Fuxin, China
Abstract—
A matching project based on SIFT arithmetic
operators is brought forward in the paper in order to realize the
matching of remote sensing images for unpiloted aircraft with
high-precision. When the unpiloted aircraft is capturing the
remote sensing images of high-resolution along the low altitude,
since the shooting angles from different stations are also
different, the geometric shapes of object imaging might occur the
aberrations and the shelter problem from the high-layer
buildings, so that results in the difficulty in matching, which
becomes the key factor affecting matching quality of remote
sensing images for unpiloted aircraft photogrammetry. Based on
the main thought of SIFT arithmetic operators by the detailed
analysis, a new matching method of remote sensing images for
unpiloted aircraft with high precision based on SIFT arithmetic
operators is put forward, and it is validated that the algorithm
has a quite strong stability by means of the actual image data for
unpiloted aircraft.
UAV; SIFT arithmetic operator; projection error; matching
Introduction )
I. INTRODUCTION
As a new type of remote sensing platform, unpiloted
aircraft photogrammetry and remote sensing system is
becoming an efficient complement form to satellite and piloted
aircraft remote sensing since it has its own idiographic
strongpoints. The chief features rest with a well flexibility, all-
weather, large area, low cost, abilities of capturing the high
precise positioning data and high-spatial resolution images in
speed velocity along the low altitude. Besides, it can be applied
in air monitoring in the dangerous regions and the area where
control surveying is not convenient to realize the map making
with non-control point.
In despite of the advantages in the remote sensing system
for unpiloted aircraft, some new problems are still brought to
the new type of earth-observation platform. When the unpiloted
aircraft flies along the low altitude to capture the remote
sensing images with high-resolution, since the shooting angles
from different stations are also different, the imaging of some
protrude ground objects, say buildings, might result in
projection error on the stereo pairs, so the geometric shapes of
object imaging might occur the aberrations and the shelter
problem from the high-layer buildings, so that results in the
difficulty in matching. Therefore, the sheltering and geometric
distortion resulting from the projection error of high buildings
certainly become the chief factor affecting the matching quality
of remote sensing images for unpiloted aircraft to this type of
photogrammetry and remote sensing platform in low altitude.
In this case, the remote sensing images matching with high-
precision is realized by using SIFT arithmetic operator for
unpiloted aircraft.
II. K
EY THOUGHT OF SIFT
A. . Extremum test in scale space
The goal of test is to primarily determine the position and
scale the key point being. The thought of scale space theory is
to carry though the scale transformation of original images by
means of Gaussian kernel so as to capture the expression
sequence of images in multi-scale space, and to extract the
scale spatial features for these sequences.
The definition of two-dimensional Gaussian kernel is
shown in Equation (1), where
σ
is the variance of Gaussian
normal distribution.
22 2
()2
2
1
(, , )
2
xy
Gxy e
σ
σ
πσ
−+
= (1)
For the two-dimensional image
(, )
xy , the scale
spatial expression in different scales is
(, , )
xy
σ
which can
be obtained from the convolution of image
(, )
xy and
Gaussian kernel
(, , )Gxy
σ
as shown in Equation (2).
(, , ) (, , ) (, )
xy Gxy Ixy
σσ
=∗ ( 2)
where, L is the scale space of the image,
(, )
y the
pixel position on image I,
σ
the factor of scale space and the
small the value, the less the image to be smoothed, the small
the scale. Large scale corresponding to the general features
978-1-4244-4131-0/09/$25.00 ©2009 IEEE