distance
erent
im
ite
(Éth)
)
50H
1H
1H
OH
so
ALGORITHMS OF DIGITAL TARGET LOCATION AND THEIR INVESTIGATIONS
À. Chibunichev
Photogrammetry Departmant of Moscow Instityte of Geodesy, Aerial Surveying and Cartografy
(MIIG AIK). Gorochovsky By-str., 4 ,103064, Moscow, Russia.
Commission V
Abstract:
Target image-coordinate measurements are caried out approximatly on am interactive digital stereo-
photogrammetry workstation by applying the measuring mark in the form of a window wich can change its
sizes. This paper deals with the algorithms of processing the windows with a target in the form af a circle, a
cross, or a square used to determine pixel coordinates of a target center. Thess algorithms are based on the
corresponding geometrical figures equations. The robust estimation technique is applied to reduce the
influence of the noise on the precision of pointing. Extensive investigations have been carried out on digital
pointing to circular, cross and square targets in order to determine the influence of quantization level, pixel
size, target size, nnise (random noise, shade, patch of light) and inclination of image on the precision of
pointing.
KEY WORDS: accuracy, algorithm, close-range, digital systems, image matching, investigation, location,
target
INTRODUCTION
Most industrial photogrammetric problems are solved
using targets of different geometrycal shape (cross,
circle ‚square, triangl and so on). Algorithms of
target image coordinate measurement in digital images
are based on determination of the center of the
corresponding figure by calculating the gravity cen-
ter (for exampele, E.M.Mikhail and al.,1984, J.C.Trin-
der.,1989, K.W Wong, H.Wel-Hsin, 1986). However,
these algorithms don’t take into account that the
image is the central projection of an object, con-
sequently all image peints are desplacad bacause af
angle het*wesn image and obiect planes. Ihesrelore,
tbe circle o n the a biect is transformed inio an ellipse
on the image,the gquilateral triangle is transformed
into a nonequilateral one and so on. In general, the
center of the figure on the image doesn’t lay on the
same perspective ray with the center of the cor-
responding figure on the object. Invariances to the
projective transformations are, as known, a paint
and a straight line. Therefore the center of a target
on the object appeares on the image in the center of
the corresponding figure only for cross and square
largets. The target center must be computed as a
intersection point of straight lines formeing a cross
or of diagonales for a square because in other wise
the gravity center will shift.
The values of these displacements (for every type of
a target) are approximately iqual to 0.5,m and lum
tor 200um and 400um of target size on digital image
(A.Chibunichev,1992). These values were calcu-
lated by means of simulation process for i - 50mm,
angles of inclination = 15, 35° 45, camera format =
5x5mm. For all shapes of targets (triangle, circle,
square and cross) ihe gravity center has been
computed which was compared with the corres-
ponding exact positions of the center. If the figure
equation is used to compute the centerof a cross and
a square, their centers coincide exactly with their
theoretical position. Thus to locate the target with
higher precision (when the angles of the camera
inclination with respect to object are large) it is
desirable: 1) to use a cross or a square as targets; 2)
to make the algorithm of target location as the
intersection of straight lines. However, as will be
shown later, it is better to use a circle as a target
when the angles of the camera inclination are not
large.
Algarithm of target lecation.
Figure 1 shows the processing steps of a semi-
automatic target location. This approach has ori-
ginally been presented in sarlier papers (A.Chi-
bunichev,1991, 1992). This algorithm serves for
targets in the form of a circle, cross, square and for
contour points which can be represented as the point
of intersection of the lines.
approximate
measurement
edge
detection
m a
> circle
lel
Cross, contour >
[=] m4
»| square »
p
matching
Fig. 1 Block diagram of targets location.
Let's consider each step of this algorithm.
l. Aproximate measurements of targets are carrisd
out applying a well known principle of stereo