Digital Signal Processing Reference
In-Depth Information
The detailed algorithm steps are described as follows:
Step 1: We can clearly see that, the center is located at the light spot nearby. An esti-
mated value can be found by searching the center of light spot with
55
template, we
define it with (X
p
, Y
p
), this template can decrease the influence of image noise.
Step 2: At the same time of thresholding, we need account the values of their gradient
directions.
Step 3: In order to improve transform speed, the binary image need be filtered again.
We define a circle that the center is (X
p
, Y
p
), the radius is 100 pixels, if the effective
pixels which are not in the place of circle will be thrown off.
Step 4: Select two effective pixels at random, the probable center can be figure out by
using previous algorithm, that is (X
o
, Y
o
). When the coordinates satisfies the condition
that:
20
X
X
Y
Y
(14)
Then apply for a memory, plus 1 in corresponding units, and record the coordinates,
find the accumulated maximum value when the operation is finished, the corresponding
coordinates are the center of concentric circles.
4
Experimental Results and Analysis
The experiment is testing the verticality of elevator track, which is evaluated by mea-
suring spot center coordinates of different locations. The image accepted by CCD
which is far from collimating instrument more than 5 meters is diffraction circles. Fig.2
shows the results of edge extraction and center position by computing the concentric
circles images at the distance of 10, 20 and 40 meters. It can be seen that there are many
pixels of ring edge at the distance of 10 and 20 meters, which can be detected easily and
accurately. The number of pixels of circle edge decreases with the increasing distance
because the edge of circle is relatively vague, which can be eliminated by median filter
algorithm, at the same time, diffraction rings will also be a corresponding reduction in
the number. Running time of detecting center coordinates is mostly less than 100ms,
which meets the requirements of quasi-real-time measurement. Shown from Fig.2, it
can be found that the pixels of circles are discontinuous, but there almost have no
influence on detection by running Gradient Hough Transform extraction algorithm.
Based on statistical thinking, the algorithm can finds out the maximum extent possible
center by same class assembling. At the same time, the effective pixels extracted at
random need not be located at the same circle, so long as the results are given to precise
positioning. Experimental data show that the quasi-real-time requirements can be
acquired, measuring accuracy is also high, which is lower than 3 pixels of experimental
measurement error.
