3D Models from Cheap Digital Images - Proposal

The purpose of this Instructable is to propose an outline for a system that can generate three dimensional models of real world objects given digital images of said objects.  In order to simplify the system we focus on generating models of faces.  However, given enough time this system could evolve to be used to generate models of arbitrary objects.  The models we aim to produce will be printable on a 3D printer, ala MakerBot.

We made this for the Things That Think class Spring 2011 at CU Boulder. We're Paul Heider , Buck Heroux and Ian Smith .

Step 1 Things You'll Need

A digital camera or a camera phone
A person whose face you want to turn into a 3D model
Our software, or possibly access to our web interface
(optional) A 3D printer

Comments

[AlternateLives]

Could this process be used to create full 3D models that can be rigged? As an animator, it would be cool to take a few pictures of a person and get a fully animated model out of it!

[paulmh]

I've never done animation, so I don't know what the requirements are for a mesh to be rigged. Unfortunately I've decided to abandon the project. It was a bit more difficult of a problem than anticipated.

[jam BD]

A proposal on how to create 3d mesh from point.

First take a plane in the 3D space and collect all the points on this plane. It is easier to just use a plane that lines up with one of the axis as that will be easier to program.

Once you have all the points on this plane use a dot to dot algorithm to generate a line along the plane. Smooth this line to eliminate jagged points in the final model. The to create the vectors for the final mesh divide the line on the plane into evenly spaced points.

Repeat this process to cover the whole model with parallel planes resulting in dots that are regularly spaced.

The next step is to join the dots that align to the plane before and after it's own into a triangle.

This process fills in the holes and will produce a smoother mesh.

As I am a programer myself I understand the difficulties. :D

[lemonie]

Publish what you achieve when you get it to work (please!)

L

[paulmh]

The focus of the project may change a little over time, but will do!

[1tri2god]

How possible is it to calibrate the software with creating a reference object within the photo that already had its dimensions inputted? So long as the reference object stays in the same location in relation to the subject, then extrinsic parameters would be accounted for.
If a reference object could not be used, I remember a seminar where this problem was approximately fixed by determining the width between pupils of the subject's eyes by inputting his height and geno-ethnicity (certain ethnicities have a standardized increased distance in relation to crown width vs. others, etc.). It wasn't a perfect fit...but much closer than no reference at all.
...just thoughts- whattayathink?

[paulmh]

I definitely like the idea of a reference object. As long as it was distinct enough from the face and is easily segmentable from the background that could work.

As for the eyes, I think it could work, but then we're back where we started if we try and extend the system beyond faces.

To be honest we hadn't though much yet about how to actually do calibration, so thanks for the input!