HOUGH TRANFORM IN DIGITAL PHOTOGRAMMETRY
C. Adamos and W. Faig
Department of Surveying Engineering
University of New Brunswick
Fredericton, N.B E3B 5A3
Canada
Commission Ill
ABSTRACT
Digital Photogrammetry requires nowadays a combination of a totally automated environment and high accuracy. Image processing,
pattern recognition, and computer vision facilitate these requirements and they are used more and more frequently. Hough transform, a
method for detecting predefined shapes, has been used for more than two decades in those areas.
The present paper proposes the use of Hough transform in Digital Photogrammetry and in particular it concentrates on the development
of a target detection technique based on the principles of this powerful transform.
Key Words: Target Detection, Hough Transform, Digital Photogrammetry.
1. INTRODUCTION
Hough Transform (HT), a method for detecting predefined
shapes, has been always recognized as a unique promising
method for shape and motion analysis in images that contain
noise, missing and extraneous data. But until recently HT was
not applied extensively due to the fact that it was
computationally expensive. During the last few years, great
progress has been achieved in overcoming this disadvantage
thus making the Hough Transform a powerful tool.
Hough Transform has been used quite frequently in the areas of
image processing, pattern recognition, and computer vision.
Some applications of the Hough Transform in these fields are
reported to be:
A system for detecting and locating mechanical parts using
generalized HT [Arbuschi, Cantoni and Musso, 1983].
Detection of straight edges of roads and tracks [Inigo et al.,
1984]. Automatic recognition of Hebrew characters [Kushnir,
Abe and Matsumoto, 1983]. Detection of faint lines in synthetic
aperture Radar(SAR) images [Skingley and Rye, 1987].
Analysis of cleavage cracks in minerals [Thomson and
Sokolowska, 1988]. Determination of motion parameters from a
sequence of images [Ballard and Kimball, 1983] [Fennema and
Thompson, 1979]. Detection and removal of linear variations
for illumination across images [Nixon, 1985]. Detection of
tumors in chest radiographs [Ballard and Sklansky, 1976]. In
recent years, HT was generalized to detect arbitrary shapes
[Ballard, 1981] and expanded to detect shapes in 3 dimensions
[Hederson and Fai, 1984].
In this paper the Hough Transform is investigated from the
point of view of the digital photogrammetry, and in particular a
target detection method based on HT is fully described.
2. GRADIENT (EDGE) IMAGE
In order to facilitate the target recognition and at the same time
reduce the amount of information presented, images are
transformed from gray level to magnitude and direction of the
gray level changes using edge operators. The most common
edge operator is the gradient operator, and the resultant image is
then referred to as gradient image.
If the image is described as a function f(x,y), where x,y are
space indicators and f the gray value for the particular pixel of
the image, then the gradient of f at (x,y) is the vector:
at
Gieyl-| 3r [es
E
250
which points at the direction of the maximum rate of increase of
the function f(x,y) (intensity).
The magnitude of the gradient G is:
G[f(x,y)] = magG = [Gx ?« Gy?]!?
and the direction of the gradient vector:
D(x,y) = tan!(Gx/Gy)
Using a 3x3 mask which is referred as Sobel operator:
X] X2 X3 -1-2-1 -10 1
X4 X5 X6 000 20 2
X7 Xg X9 3 241 -10 1
Gy and Gy for the center pixel of the mask in the above
equations are given by:
Gx = ( x7+2*xg+X9)-(X,+2*X3+X3)
Gy = ( x3+2*x6+X9)-(X1+2*X4+X7)
Setting a threshold in the magnitude of the gradient, every pixel
that exceeds this threshold is mapped as edge point, otherwise it
is mapped as background (zero gray value). In this way the edge
image is generated and each pixel is assigned with the magnitude
and direction of the gradient vector which will be used in
forthcoming tasks.
3. HOUGH TRANSFORM FOR DETECTING CURVES
The classical Hough Transform is a curve detection technique
that can be applied if little is known about the location of a
boundary but its shape can be described as a parametric curve
[Ballard and Brown, 1982]. Furthermore, Hough Transform
has been expanded to detect arbitrary shapes [Ballard, 1981].
The method is attractive because it is quite unaffected by gaps,
partial occlusion and noise. Hough Transform is regarded as an
intermediate level vision task since it converts an iconic
represi
1989].
[Duba
Transf
approa
where
directi
1). Thi
size ( c
0° anc
Then a
parame
into sin
of imag
parame
these pc
in the i
The con
from su
A(p,8)
Fig.