I’m
agnostic when it comes to 3D machine vision. I see the value in
acquiring another channel of data but I’m concerned about the
complexity of processing it. So I read Winn Hardin’s “3D
Vision Thinks Outside the Envelope”,
(Visiononline.org, May 17th,
2013,) very carefully.
I
didn’t learn a whole lot – Cognex say their 3D camera and
VisionPro tools are great, LMI think their Gocator is the bees-knees
– but I did follow a link to the Ensenso
N10 stereo 3D camera,
sold by iDS.
After
satisfying myself that it’s not a rebadged Point
Grey Bumblebee,
I wondered about the third port midway between the two camera lenses.
Turns out, this is where they project a structured light pattern
from. It’s the technique at the heart of “Projected Texture
Stereo Vision” which is what makes stereo a useable tool.
Don’t
feel bad if that phrase is unfamiliar to you, I hadn’t heard of it
either, but Google has. A paper titled “Projected
Texture Stereo”
and put out by Willow Garage explains the whole thing. Math isn’t
my strong suit, but the bottom line is straightforward: projecting a
pattern onto an image provides the system with a whole lot more data
from which to compute a disparity map. In other words, more data
equals a better result.
1 comment:
The idea is actually really simple. You want to use stereo to reconstruct parts of the scene which have no texture, so you artificially create a texture through projection.
Post a Comment