Dynamic Projection Mapping Onto Deforming Non-Rigid Surfaces

H

HardOCP News

[H] News
Joined
Dec 31, 1969
Messages
0
You guys have to watch this video. What you are seeing is a high-speed projector that can project 8-bit images up to 1,000 fps with 3 ms delay and compensate for warped and moving surfaces. Is that cool or what?
 
They essientially do the same thing to calibration vision for robotics. I don't think this is that hard to do...
 
super hi res/fast version of MS-Kinect that measures the change in distance between points. It's the point mapping that is most impressive here. They would have to use some sort of matrix lookup to determine likely movement vector assuming a cord L of non changing length.

This will allow network stations to dynamically map new graphics on top of objects they want to hide, or to show advertisements on. This was done before with static objects in stadiums, but dynamic movement is quite complex.
 
That was the opposite of what I thought it would be. I thought it would make an image look flat on an irregular surface. Oh well...
 
DigitalGriffin said:
super hi res/fast version of MS-Kinect that measures the change in distance between points. It's the point mapping that is most impressive here. They would have to use some sort of matrix lookup to determine likely movement vector assuming a cord L of non changing length.

This will allow network stations to dynamically map new graphics on top of objects they want to hide, or to show advertisements on. This was done before with static objects in stadiums, but dynamic movement is quite complex.
Click to expand...

I don't quite exactly know what a matrix look up is but it definitely appears to be doing 3D transformations of the projected image using the IR markers as 'resolution' points. No matter what the method, processing any kind of real-time transformations like that in 1,000 fps is impressive.
 
  • Like
Reactions: raz-0
like this
Dayaks said:
They essientially do the same thing to calibration vision for robotics. I don't think this is that hard to do...
Click to expand...
So they are selling aiming assistance technology as a funny gimmick so Robots can better kill us doing the uprising ?
 
gunbust3r said:
You're reading too much into it. These guys are just really into anime and want a virtual outfit for their favorite j-pop singer. (Or pillowcase)
Click to expand...

That's a fucking racist stereotype man, those people are just as entitled to dignity as you are, and they have a much harder life than you; fighting godzilla and having tentacle sex.
 
Sonicks said:
I don't quite exactly know what a matrix look up is but it definitely appears to be doing 3D transformations of the projected image using the IR markers as 'resolution' points. No matter what the method, processing any kind of real-time transformations like that in 1,000 fps is impressive.
Click to expand...

You would have to assume that the fabric or whatever material has a fixed length "L" Based on the x,y position from the point mapping, you can calculate the z if they are using ir markers. If they are using a system like kinect, they can read the Z (depth) and figure out x, and y based on that fixed length. Length = integral of F(x). of the derivative of the area evaluated at the boundaries. (sort of like how the derivative of a circle total area gives you the surface area of a circle) Any way, you want to create the illusion of a flowing point to point smooth movement as fabric does. The fastest way is to use cubic spline or interpolations. And you want to to it in such a way that the cubic spline from time t is similar to t+1 to create smooth flowing movement. This has been done in fabric simulations multiple times, and is extremely GPU intensive. It's likely easier in a simulation as you know what forces are being applied to the fabric to predict it's movement. Although it might be possible to offload it all onto a GPU, a pre calculated movement solution might be a more processor efficient way to handle it. And that pre-calculated is based on a matrix lookup. Given the high sampling rate they are likely trying to more accurately predict the movement vector so the lookup doesn't have to make large interpolation guesses which are harder to correct for if wrong.

Any way, that's my worthless hypothesis of the day.
 
You must log in or register to reply here.
Back
Top