DoubleTake

DoubleTake is a project I worked on in my spare time from August to September 1996. It's a tool to help create 3d models from 2d photographs. I put it on the shelf when I realized I didn't know how to automatically deduce the position of the camera for each image. I might start working on it again sometime.

Anyway, this page should give you an idea of how it works.

1) get two or more photographs of the same object, taken from different positions.

2) mark "features" on the images.

3) correlate the features.

I can't show this in pictures. To do it, you click on a feature in one image and then on the "corresponding" feature in the other image. Now when you move the mouse over a feature, it highlights itself and all the features it's correlated with. 

4) connect the dots.

5) export to a VRML file.

Goals

• simplicity: it should take less than half an hour to become an expert. 
• speed: 20-30 features per minute. An average vertex will use 2-4 features, so that should make 5-15 vertices per minute.
  It took me about 75 seconds to create the model shown above. That's about 10 features a minute. It would have been faster if I hadn't counted the time to open files, resize windows, and save the model to disk. The zoom feature probably would have made it faster as well. On the other hand, it would have been slower if the model had been more complicated than a box.

To Do

• add texture mapping (currently every model is a boring grey)
• get some more interesting photographs
• make a demo version that works inside a WWW browser (it's written in Java, after all)
• add a zoom feature to make positioning features easier
• correct for lens distortion (once I start using real cameras)
• figure out the position of the cameras automatically

Last updated September 21, 1996
Back to Kimberley's Code.
Back to Kimberley's Home Page.