Relay-Version: version B 2.10 5/3/83; site utzoo.UUCP
Path: utzoo!utgpu!water!watnot!watmath!clyde!rutgers!mit-eddie!uw-beaver!ssc-vax!thornton
From: thornton@ssc-vax.UUCP
Newsgroups: comp.graphics,comp.ai
Subject: Re: Solid Modeling
Message-ID: <802@ssc-bee.ssc-vax.UUCP>
Date: Sun, 29-Mar-87 15:52:37 EST
Article-I.D.: ssc-bee.802
Posted: Sun Mar 29 15:52:37 1987
Date-Received: Wed, 1-Apr-87 06:59:53 EST
References: <798@ssc-bee.ssc-vax.UUCP> <1055@rpics.RPI.EDU>
Organization: Boeing, Seattle WA
Lines: 79
Keywords: 3D Graphics, Robotics, Automation
Xref: utgpu comp.graphics:424 comp.ai:346

In article <1055@rpics.RPI.EDU>, chassin@rpics.RPI.EDU (Dave Chassin) writes:

> I think you have another problem when you get involved with robotic vision,
> and this is something that I've never thought about in terms of robotics, but
> I am working on in terms of architectonics (architectural modeling of sorts).
> That is that you will need to create some sort of algorithm for generating a
> 3D model from 2D information received by the cameras. Essentially the idea
> is to analyse a pair of images, extract the boundary data, assemble a 2D
> 'image' for each view, project the two images together into a 3D 'image',
> and finally take the resulting B-rep data and convert it to CSG type data,
> which can then be correlated with the previous frame and the motor algorithms
> to properly direct the parts into their desired positions. Piece of cake, eh..
> 
> Each of these steps involve some very complicated and SLOW computing. I've
> worked out the basics for the first 4 steps, but have a long way to go
> still. In any case I would love to talk more about the ins and outs of this
> type of analysis because this is the main focus of my work for the next
> year or so. By the way, it's all being done on a Sun 2/120 and 2 AT clones...
> ...wish me luck!!!

From what I understand, you are using solid modeling for architectural
design, but may also wish to construct models from existing designs.

> I know there are some people who have already done some work in these areas,
> but it has always amazed me how little is in fact published. 
 
I consider these issues ones of object representation and object
reconstruction. The reconstruction problem is very difficult, especially
for architectural designs. One source that I have found that discusses some
of these issues is "Three-Dimensional Object Recognition" by Besl and
Jain, ACM Computing Surveys, Vol. 17, No. 1, March 1985. 

Unfortunately, most of the work on model construction seems to be based
on lines drawings. This is probably due to the difficulty in obtaining
accurate depth data from an image. Having accurate surface normals sure
would help the situation. I know some work how been done with range data,
but I don't know how accurate or sparse these measurements are.

My particular interest, at least now, is determing what information the
solid modeling system can provide that can be used for visual inspection.
The solid modeling system and vision systems obviously both use geometrical
information about the part and need algorithms for displaying the part.
In addition, as I mentioned earlier, the vision system needs information
about surface reflectances and color as well. For this reason, it might
be best to integrate parts of the vision system with the solid modeling
system, or at least be able to share some of the core software.

This brings me full circle to my original query. What systems are people
using? I know that constructive solid geometry and boundary representation
seem to be the most popular, but what about spatial-occupancy (voxel or
hyperpatch) representations. Voxel representations seem to be extremely
memory intensive but, as greater demands are made on the complexity of
required models, will these representations come of age?

Finally, the only solid-modeling system that I'm aware of, for which
source code is availabe, is PADL-2, born at U Rochester. Also, I 
understand that it is written is FORTRAN. Has anyone re-written this
system in C and successfully ported it to a unix machine, such as a 
Sun 2/3? If it is to be used for university research, what's the cost
of PADL-2? Are there any other systems for which source code is 
available? (Note: although my signature doesn't indicate it, my work
is being done in conjunction with the U of Washington in Seattle, so
I expect we could get an educational discount.)

E-mail replies are greatly appreciated, but I am interested in generating
network discussion on many of these issues, if the discussion is deemed
appropriate. 







-- 
     /\          
  /\/  \/\   
 / / /\/  \ Ken Thornton  {decvax,ihnp4}!uw-beaver!ssc-vax!ssc-bee!thornton   
/ / /  \   \