English subtitles

← 17-32 Correspondence Issues

Get Embed Code
2 Languages

Showing Revision 1 created 11/28/2012 by Amara Bot.

  1. This segment is my explanation of correspondence in stereo vision.
  2. It came a long way. There are a few things that don't work really well.
  3. For example, we have two cameras over here, and we have a big object over here
  4. with a foreground separate object.
  5. Then the order contraint is being opposed and dynamic programming doesn't hold.
  6. That is, an object over here might appear left of the object over here in the left imager
  7. but right of the object over here in the right imager.
  8. There are other cases where things go wrong.
  9. For example, suppose you were imaging a circular object with these two imagers here.
  10. Then the occlusion boundary of this object as viewed from the right imager
  11. is different from the occlusion boundary of the same object as viewed from the left imager.
  12. These are not corresponding points. They correspond to different points on the object.
  13. As a result, your stereo calculation will give you a poor result.
  14. A final instance where things might go wrong is
  15. reflective objects that have specular reflections.
  16. This ball over here reflects the ceiling lights,
  17. and obviously, where the ceiling lights are being reflected
  18. is a function of where an imager is positioned.
  19. For these specific features over here,
  20. we get a really lousy depth estimate for the object at hand.