Submitted version of the thesis - Airlab, the Artificial Intelligence ...

5.3. Tracking 65
any color values. It means, if the object is exposed to light, the parts that
are exposed to light are not considered as white but a tone of target color,
similarly with the previous configuration the dark pixel such as the parts in
the shadow are considered as black, but with the new configuration they are
also considered as a tone of the target color.
Second step is calculation of the color information in a better way. Previously,
we calculated the information by ST color picker software. With
this software we select the color region we want to assume as the blob color.
The problem of this approach is that the blob may not consists of a single
rule (RGB color code). Some parts of the object are darker, some parts
are lighter. Also with the distance, the color that is composing the object
is changing. Additionally, the colors that are not considered as red, blue,
green, represented again as rules (RGB color codes) and it is not possible to
understand by just looking at the color code whether it is the correct color
coding or not. These makes it very difficult to determine the color with the
ST color picker software. Instead, we decided to find the color histogram
that is the distribution of the selected color properly in the RGB channels
separately. To do so, we took several samples from the camera, in different
distances, in different illuminations to find a proper average. The calculated
results are later tested with the ST color picker software, since it offers a
good interface to see whether the color is selected correct or not, in a visual
way.
The third step is related to the estimation of the objects distance to the
robot. As mentioned before, the diameter of the ball is a problematic issue
due to the low-cost optics, hence we decided to use an object with cylindric
shape, and the second configuration for the transformation. This distance
is easier to measure, independently from the shape of the object, and will
always change as the object’s position changes in the world.
During the implementation to the microcontroller to increase the performance,
we tried to avoid matrix and floating point operations. The matrices
that can be defined statically are defined statically. To avoid floating point
operations, the floating points are converted into integer, then multiplied by
a constant (most of the times 1000). The operations are made in integer,
and the result is returned as floating point by dividing with the constant.
We ensure that the microcontroller will not spend too much time in matrix
and floating point operations, using this approach.