-
Notifications
You must be signed in to change notification settings - Fork 111
team configuration
SSL-Vision's supplied pattern detector should be able to support any arbitrary
color-based multi-segment pattern as typically used in the Small Size League. A
team's patterns are defined through a .png file in the directory
ssl-vision/patterns. For official competitions, the patterns are standardized. The standard patterns are defined in patterns/teams/standard2010.png. This
team image contains a grid of individual robot covers, and it is parsed in
row-major order, with increasing robot IDs, starting at 0. This image will then
be automatically parsed, and the pattern detector will use the areas and
relative angles of all markers defined in this image for pattern detection. For
the image to be parseable, each robot should contain a blue center marker.
Furthermore, as shown in the example image, each cover image should be
accompanied by a single pixel with yellow vertical bar which indicates the
height of the robot with its height in pixels corresponding to the robot's
height in mm.
Once this pattern image has been defined, it is possible to create a new
team-definition through the GUI. To do so, open the branch Global/Team Config/ in the Data-Tree. Next, click on Add Team.... This should
create a new entry in the Global/Team Config/Teams branch labeled New Team n. After opening this team's branch, you can rename the team as desired.
To make use of the png file, make sure to check "Unique Patterns" and "Have
Angles" in the team's config branch. Then open the branch "Marker Image", make
sure "Load Marker Image" is checked, and enter the path to the png file as seen
from the ssl-vision main directory. For example:
patterns/teams/CMDragons.png. Next, make sure that the number of robot
rows and columns are correct. Finally, uncheck any of the robot IDs which are
not being used in your robot definition file, in the "Valid Patterns" field.
To use your new team pattern definition effectively, some parameters might need to be tweaked. Each team definition is able to carry a unique set of parameters as follows:
This defines the parameters of the blob-filter used to detect the Blue or Yellow center markers. Most of these parameters should be self-explanatory, except the following ones:
-
Uniform, Exp. Area Mean, Exp. Area StdDev. are all used to compute the
confidence of the marker based on its measured deviation from its expected
area. The confidence is computed using the following formula:
conf = gaussian(measured_area_deviation, expected_area_std_deviation) / (gaussian(measured_area_std_deviation, expected_area_std_deviation) + Uniform)wheremeasured_arae_std_deviationis the absolute difference between the markers measured area and "Exp. Area Mean". - Duplicate Merge Distance means the maximum distance where two detected center blobs should be merged into one.
This defines the parameters of the blob-filter for all surrounding markers (i.e. not the center marker). All of these should be self-explanatory.
Note that the Histogram is currently only used for detecting Unknown teams (i.e. teams without a marker image)
The Histogram checker validates whether the detected blue or yellow center markers (of team's for which the surrounding patterns are unknown) are plausible to belong to an actual robot. It does so by computing a color-histogram of surrounding pixels to the marker which allows to check for the "markeryness", "field-greenness", and "black/whiteness". Markeryness is defined as the ratio of the total number of pink,bright green,and cyan pixels over the number of pixels colored in the color of the corresponding center marker (blue or yellow). Field-Greenness is the ratio of the number of dark green pixels over the number of all pixels in the histogram. Black/Whiteness is the ratio of all black, clear, and white pixels over the number of total pixels in the histogram.
These settings control the evaluation of the actual pattern matching algorithm which uses the pattern definitions as previously defined in the team's .png file.
- Max Marker Center Dist is the maximum distance of any surrounding pattern's centroid from the central team marker's centroid.
- Weight Area is the weight of the "area matching" error component (i.e. how well do the detected areas match the ones in the pattern definition)
- Weight Center-Dist is the weight of the "distance from center marker to surrounding markers" error component
- Weight Next-Dist is the weight of the "distance from one surrounding marker to the next" error component
- Max Error is the maximum allowed total weighted error of the three components described above, to count as a fully confident detection.
- Expected StdDev is the expected deviation on the error
- Uniform is the uniform component against which the error and deviation are compared, similar as described to the formula further above in the "Center Marker Settings" section.
Once you have fully configured your team, you can select it under the
Global/Blue Team and Global/Yellow Team branches.