Installation manual
308
can copy and paste the proper pixel values from this sample image to construct a
new image, one pixel or line at a time.
This is the Target-T.bmp file (...\setups\CustomApps\Feedback\target-t.bmp):
• Note that the FOREGROUND is the T shape, except that the intersection point
is in the REFERENCE_POINT color.
• Not
e that the background markers are single pixel lines as well. This is for
performance. Adequate recognition will occur using single pixel lines for the
foreground and background, and the fewer pixels we have to analyze the
better for speed.
• There is no need t
o restrict yourself to single pixel wide lines. Nor is there a
need to be restricted to orthogonal lines. For example, you could design a Y
pattern, L pattern, etc.
• B
e careful not to design patterns that are impossible to match for different
camera zoom levels. The camera placement on current vision modules is
inconsistent, so images will appear different size on different laser modules.
You wouldn't want a closed shape in such a case, because shapes will be
different sizes on different cameras at different times, due to camera
positioning discrepancies.
• Open targets are suitable. These are targets that have an intersection of
straight (but not necessarily orthogonal) lines. The lines need not intersect.
• The target
shape should be chosen so that it will not be easily confused with
other artifacts on the background of the image. Keep in mind that as products
are inserted in different orientations, you want to be able to reject misoriented
products (not find false matches).
• Ta
rgets can typically be quite small and yield excellent search results, if they
are carefully designed so that there aren't “false matches” elsewhere on the
product. For example, the Target-T.bmp sample is only 30x30 pixels in size,
which comes to about 1mm by 1mm for the laser module. A target on the
product can be this small and still be reliably recognized.
• Not
e that the targets for the Ring Search algorithm can often be much smaller
than targets for the ConVision algorithm, and still yield excellent results.