Computer Vision Issues

The Celestial Curling prototype was set up yesterday and was confronted with a number of Computer Vision issues. Aside from the embarrassingly long setup time, the camera image gave back a horizontally-inverted image (a quick fix with the math in Processing) and the camera had an awful lot of difficulty picking up game pieces that weren’t directly in the middle of the screen. In order to address this, I cracked open the case of the camera, and mounted it higher up, leaving room for a paper cone construction to diffuse the infrared illuminators light outwards. The paper seemed to block too much light, although the small amount of light that it put out seemed evenly distributed. Operating in these low-light conditions will likely create a lot of noise in the computer vision, so I’m going to do a few things that should help.

  • use mirrored/shiny tape on the topside of the illuminator, instead of the reflective tape currently used. This should hopefully help light up the edges of the playfield more evenly.
  • construct a taurus-shaped velum diffuser for the light that should be more translucent than the paper cone i tried.
  • make one last attempt to get Intel’s openCV library doing the computer vision for me. OpenCV contains a lot of code to track the motion of objects, regardless of color that the JMyron library I’m currently using lacks. I figure this would save me in case I’m unable to get the IR light levels sufficiently high.
  • lastly, I am going to add a bunch more IR led’s to the illuminator to pump up the light levels.

Below is a video of my object tracking code working without the acrylic playfield in place.

Comments are closed.