• Welcome to Redshift Project Depot.

Vision system

Started by Louis L, January 24, 2016, 08:08:16 PM

Previous topic - Next topic

Louis L

I played around with the LabVIEW vision sample VI. I tried both the sample image files (the same as used on GRIP) as well images captured from the Microsoft camera. I want to try capturing from the Axis camera next.

Some random observations:

  • The MS camera was picking up a lot of noise from the room - lots of surfaces in the workroom reflect green light.
  • The distance calculation is off; we used a tape measure and it didn't add up. Rakesh and I looked at the code and it didn't look right. We'll have to investigate further and fix it.
  • By the same token, other metrics don't seem right either - like the "moment" calculation.
  • What it does - it take a image frame, puts the image through a color threshold filter to yield a binary bitmap. It then perform a particle analysis to create a list of hits. The shape it's looking for is a convex hull. Finally it calculates an average of some sort and sorts that list from best to worse. There's probably a lot more to it (it's not trivial code) but I haven't had time to dig further.
  • One interesting tidbit - the vertical vision retro tape is only 12 inches tall - half the height of the goal's opening. So after the target is acquired, the real target is the middle of the upper bound of the resulting parallelogram since that defines the mid-line of the goal.
  • The mount on the Microsoft camera is overly adjustable. We need it to be very stiff so we can rely on it despite bumps and bruises. This needs an ECO.
  • The angle of view on the Microsoft camera is more narrow than I expected. Need to compare it to the Axis. The issue here is that the shooter does not elevate so we need as wide a vertical coverage as possible so we can still see the upper goal from close-in to out-far. I've asked Ryan to think about possibly adding a servo for elevation control to just the camera. Another option is to rotate it 90 degrees for sideways targeting but that is best done at the point of rotation (when rotating in-place).

Louis L

I think I mentioned this elsewhere but we decided go with the Axis camera over the Microsoft camera. The viewing angles are about the same - the Microsoft may be a tad wider. We'll send the video to the driver station and also process it on the roboRio.