Another idea (in addition to the strong color background and kill color function) for parts recognizing (from OpenPnP group): “glass with soot from burnt wood on the backside. The part sits on the clean glass surface”
https://groups.google.com/forum/m/#!top ... 9caHpS7uoA
Another idea for parts recognition
Re: Another idea for parts recognition
Where is the advantage vs black cardboard or black foam mat?
Thinking of the reflexion of the down light on the glass surface you will get bright spots that run your elsewise perfect black hole.
I am currently testing an other approach having the parts on the white desk top.
Then using edge detection after a selective threshold and then detecting contours with bounding rectangles you get quite nice center points (of the bounding rects) to pickup a single part.
By setting a filter on the bounded rectangle size you can even distinguish LEDs from Cs and Rs.
If the parts are not too close together you get a list of center locations.
If you need to align the part you can then place it to a spot where only the part is visible, zoom in and use a method similar to jason niedas rotated rectangle for his up cam alignment (although this method afaik still has a minor angle recognition problem that seem to be solved here https://stackoverflow.com/questions/240 ... nformation)
I am using jasons method (without his color filtering, he is using to filter the visible nozzle in the background) with my down cam to look into the tape pocket of 5050 LEDs which tend to be slightly rotated due to oversized pockets.
But instead of calculating offset and rotation I iteratively center the camera like the optical homing does until the camera/nozzle are direct above the center, and use the rotated rects angle to turn the nozzle A to compensate the parts rotation in the pocket. Then I pick up the part and go back to the tapes part orientation angle or to the parts angle on the PCB and place it.
Thinking of the reflexion of the down light on the glass surface you will get bright spots that run your elsewise perfect black hole.
I am currently testing an other approach having the parts on the white desk top.
Then using edge detection after a selective threshold and then detecting contours with bounding rectangles you get quite nice center points (of the bounding rects) to pickup a single part.
By setting a filter on the bounded rectangle size you can even distinguish LEDs from Cs and Rs.
If the parts are not too close together you get a list of center locations.
If you need to align the part you can then place it to a spot where only the part is visible, zoom in and use a method similar to jason niedas rotated rectangle for his up cam alignment (although this method afaik still has a minor angle recognition problem that seem to be solved here https://stackoverflow.com/questions/240 ... nformation)
I am using jasons method (without his color filtering, he is using to filter the visible nozzle in the background) with my down cam to look into the tape pocket of 5050 LEDs which tend to be slightly rotated due to oversized pockets.
But instead of calculating offset and rotation I iteratively center the camera like the optical homing does until the camera/nozzle are direct above the center, and use the rotated rects angle to turn the nozzle A to compensate the parts rotation in the pocket. Then I pick up the part and go back to the tapes part orientation angle or to the parts angle on the PCB and place it.
best regards
Manfred
Manfred