Relay-Version: version B 2.10 5/3/83; site utzoo.UUCP
Path: utzoo!utgpu!water!watmath!clyde!rutgers!husc6!ut-sally!utah-cs!utah-gr!bogart
From: bogart@utah-gr.UUCP
Newsgroups: comp.graphics
Subject: View Correlation
Message-ID: <2234@utah-gr.UUCP>
Date: Wed, 18-Nov-87 00:26:20 EST
Article-I.D.: utah-gr.2234
Posted: Wed Nov 18 00:26:20 1987
Date-Received: Fri, 20-Nov-87 23:37:22 EST
Organization: University of Utah CS Dept
Lines: 26

I am attempting to solve a view correlation problem.  Given a set of
3D points and the 2D coords of each from a photo image, find all the
parameters necessary to render (or raytrace) the points and get the same 2D
coordinates.

For example, I have a photo of an object which I know to be a cube.  I
can assign 3D coords to the corners of the box, then find the
locations of each corner in the image.  I would like to throw all of
that data at a black box, and get back the viewpoint, view direction,
aspect ratio, etc.

So far, I have tried solving this using methods from Rogers
"Mathematical Elements for Computer Graphics" and various Utah theses.
All of the solutions seem to very unstable numerically, i.e. if the
the points are ordered differently, I get a different solution.  This
is because these methods attempt to solve a system of equations that
is rather large, and the math breaks down if there is error in the
original data.  I suspect that an iterative technique would be more
suitable, but I have no references for one.

Any help out there?  I would greatly appreciate mail from anyone who
has actually written code to solve the problem, but feel free to send any
references, suggestions or ideas that seem appropriate.

RGB
(Yeah, those are my real initials...)  bogart@cs.utah.edu