2. THE METHOD EVALUATION
This section presents an evaluation of the method and gives the
practical results of the experiment. As mentioned in Section
one, in the initial position the visual system acquires a stereo
image; whereafter, the program automatically constructs a
mathematical stereo model as explained in Homainejad and
Shortis (1995b). Then the visual system is relocated in a new
position; whereafter, it acquires some stereo images whilst a
dynamic object is moving throughout the scene from the left
side to the right side.
The direction. : : Test
of the i field
movement of :
the object.
nl The first and the
second positions
p ofthe visual
| | zz | | system.
Figure 1: The chart of the movement of the visual system.
2.1 Program Evaluation
At the first step, the program defines the position of both
cameras. In this stage, the program uses some constraints that
are introduced to the computation. The constraints address the
approximate direction of the movement of the visual system.
These constraints do not limit the all the movements of the
visual system; ie, the base line has some free movement in the
visual system. After replacing the visual system, the base line is
rotated around the Z and X axis. As a result, the program
should define the position of the two cameras separately, rather
than defining it only for one camera. The parameters of exterior
orientation for the two cameras are presented in table 1.
Table 1: Parameters of exterior orientation for the two cameras.
If the visual system has no freedom of movement and its
baseline is imposed always perpendicular to the Z axis and
coincident with the X axis, then, defining the position of one
camera is essential and the position of the second camera can
simply be defined according to the position of the first camera.
Positioning two cameras increases the time delay by at least 10
ms. The program registers the three parts of the image (that
includes the three templates) into three arrays. This processing
reduces the time delay by 50 ms. The longest time delay in the
processing is related to the retrieval of the image into an array.
It should be noted that for fast and precise processing only three
control points are needed by the program. The constraints
enable the program to recognise the approximate position of the
three templates in the images. The reduction in delay time
depends on the size of the image; ie, a larger image has a longer
244
time delay. The size of the images in this experiment is
768x580 pixels. It should be noted that the processing is carried
out on a SUN SPARC 20 work station. The images are in TIFF
format, and the TIFF library was used for image processing.
Other methods for restoring and reading the images were tested
but were slower than the TIFF library. For example, reading
and restoring TIFF images using the C library is 150 times
slower than the TIFF library; therefore, it is preferred to use the
TIFF library to be used in this experiment.
In programming, Loop is another characteristic that increases
the time delay. The processing of over 20000 characters in the
program are common for positioning of the left and the right
cameras. In order to process these 20000 common characters,
four approaches are tested. As the first approach, a goto
command was used for the processing, but the time delay was
more than 100 ms. The goto command was replaced by the
while command in the second attempt, and the time delay
improved by 40 ms. In the third attempt, the for command was
tested and the time delay improved by 2 ms. In the fourth
attempt, the loop was removed and in conclusion, 20000
characters are increased to the program. As a result, the time
delay for the positioning of the two cameras and stereo
matching was improved to 30 ms in the fourth approach. The
process of the positioning and stereo matching is pseudocoded
in Figure 2.
while ( The condition is true ) (
for ( i=0;i<nji++){
Retrieve the part of image and restore into
the array according to the constraints.
Measure the coordinates of the coordinates
of control points and read their number.
}
Compute the position of the camera and the
external parameters.
if (rigth image ) break;
}
Compute stereo matching.
Figure 2a: The pseudocode for the second approach.
if (left image ) {
for ( i=0;i<n;i++){
Retrieve the part of the image and restore
into the array according to the constraints.
Measure the coordinates of the control
points and read their number.
}
Compute the position of the left camera and
exterior parameters.
}
if ( right image ) {
for ( i=0;i<n;i++){
Retrieve the part of the image and restore
into the array according to the constraints.
Measure the coordinates of the control
points and read their number.
}
Compute the position of the right camera
and exterior parameters.
}
Compute stereo matching.
Figure 2b: The pseudocode for the fourth approach.
Table 2 compares the time delay for the four approaches. The
result is reliable, robust and precise. It is reliable and robust
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B5. Vienna 1996
because the
recognises a «
the control p
because the [
points with
intersection m
object system
RMS error of
pixel. Table 3
computed cor
Table 3: Th:
coordinate
Once the pri
matching are
dynamic obj
According to
shape or its «
deform. The
waist. The w:
in the test fie]
Figur:
The strategy
experiment. ,
à point upon
the right imz
experiment i
background
point in the
images and
extracting th
pattern in the
program. Th
of a commor
the weight a
image, the ai
