Page 2 of 2

Re: rmod (ver2) help please

Posted: Mon Jun 13, 2016 3:48 pm
by vonnieda
Pixopax wrote:Jason, I saw that the opencv gaussian blur function makes my framerate go down very strong, do you know if that is opencv-related?
I have a very fast computer, but as soon as I use gaussian blur I only have a fps of 1-2 frames / second left.
Guassian blur is indeed very slow, and you really only want to do it during vision processing, not for every frame from the camera.

Re: rmod (ver2) help please

Posted: Tue Jun 14, 2016 8:03 am
by Pixopax
But it makes detection better, it eliminates small reflections from the plastic tape.
I will see if I can change the algorithm in only using it on a single frame.

I tried that filter out on a clear tape which i made black with a marker. That makes detection much easier, but the light reflects heavy as it is a high-gloss plastic. Blurring this makes it better, but is very slow.
Now I look for a way to make these clear tapes black with a dull finish, that would be perfect.
I could not sonsistently place parts from clear tapes, and almost all larger Cs and LEDs come in clear tape.

Re: rmod (ver2) help please

Posted: Tue Jun 14, 2016 11:10 am
by JuKu
The measurement algorithm in standard software does that (takes single frame for processing and measurement), not sure what r-mod does. I'll see about adding blur.

Re: rmod (ver2) help please

Posted: Tue Jun 14, 2016 12:11 pm
by mrandt
I am using 3D printed strip feeders from either blue or green material; so I can always detect the sprocket holes no matter the colour of the tape.

I have shared the design and STLs here:
http://malte-randt.de/strip-feeders-and ... y-for-smd/

Regarding reflections on tape I believe these are mostly caused by inadequate lighting; so add some more blurring to your light source and / or adjust brightness (with high speed PWM).

In addition to that, I think that sprocket hole detection is nice but in many cases it would be more reliable if LitePlacer also supported a much simpler approach:

I wish I could teach the software the center of first and last component using camera and jogging controls. By setting tape width and component pitch / component count, it could then easily do the maths and figure out the location of each part...

I documented that feature request in November last year:
https://github.com/jkuusama/LitePlacer-DEV/issues/29