Thanks for simplifying and Great breakdown of the vision system algorithm using openCV computer vision for component alignment in this video shorts

Hast du den Algorithmus für die Winkelerkennung selbst geschrieben, das muss ja auch digitalisiert und verglichen werden, brutal... echter respekt!

Hi Kyryl, the resolution of the camera is currently 0.07mm per pixel. So this is the overall limiting factor. The rotation accuracy will then be a function of the part size. I have redesigned the algorithm to do multiple alignment steps until the part is aligned to X0 and Y0 +/-0.05mm and rotation is 0 degree +/- 0.02 degree. These are the output values from the bounding rectangle detection not including the overall limits mentioned above.

Is there any provision to visually align the BGA components with the PCB foot print..?

Pretty cool. Have seen other DIY and most did not appear to have a visual aligning of the part. Just what appeared to be valid aligning of parts in the reels. On ic' show did you verify pin one placement. I'm new to all of this and all I've seen is that the top of the ic has the indicator. Pretty cool though.

Not bad, to say the least! What is limiting your accuracy/precision in alignment? Hope it's not vision algorithms but rather servo step size and backlash? Also, on rotation alignment, what is the best accuracy you've been able to achieve so far? In the video it looks like ±0.5° which is insufficient for large QFPs...

Forgot to mention, the output of the bounding rectangle algorithm is averaged over n samples.

It is a stepper motor from nanotec.

Can you explain the principle of work of the software in details, please? Can we try this software - where download?

@VBsProjects

10 жыл бұрын

You can download the software - including the source code and a lot of documentation - from my website that is linked into the description.

Hanzhen harmonic drive gear , over 30 years experience , strain wave reducer, robot joint ,

I guess I should give OpenCV a try again. In my first attempts the MinAreaRect was not as robust as I wanted it to be on asymmetrical parts and I resorted to using a commercial vision library instead. I'm sure you've seen my first test here: watch?v=9bODRom1jrs