Date: 06 Aug 93 11:28:17-PST
From: Vision-List moderator Phil Kahn <Vision-List-Request@TELEOS.COM>
Errors-to: Vision-List-Errors@TELEOS.COM
Reply-to: Vision-List@TELEOS.COM
Subject: VISION-LIST digest 12.35
To: Vision-List@TELEOS.COM

VISION-LIST Digest    Fri Aug 06 11:28:17 PDT 93     Volume 12 : Issue 35

 - ***** The Vision List has changed hosts to TELEOS.COM *****
 - Send submissions to Vision-List@TELEOS.COM
 - Vision List Digest available via COMP.AI.VISION newsgroup
 - If you don't have access to COMP.AI.VISION, request list 
   membership to Vision-List-Request@TELEOS.COM
 - Access Vision List Archives via anonymous ftp to FTP.TELEOS.COM

Today's Topics:

 Re:  A Hough Transform Program
 IMSL routines in the Tsai article
 A frame grabber device driver for Linux
 3D object recognition using affine invariants
 CFP: Toward Physical Interaction and Manipulation
 Ukrainian work

----------------------------------------------------------------------

Date: Thu, 5 Aug 93 14:00:30 EDT
From: herrell@cps.msu.edu
Subject: Re:  A Hough Transform Program

[ The following Hough Transform PD code has been placed in the SHAREWARE
  subdirectory of the Vision List Archive. Thanks to Richard!
						phil...		]

I have recently written a simple program that accepts an
image in PGM (Portable Grey Map: A VERY simple format )
format and outputs one of:

1) a pgm formated image that is the hough transform of the given image 
   (standard line detection r, theta space)  

2) A list of r theta confidence, r theta confidence ....
   for all of the local maximums.

3) A pgm formated image where all non local maximums in
   the hough space are blacked out.

depending on the command line options given.

I have also written a separate program which accepts
the original pgm formated image and the output of the
hough program and outputs a ppm ( color ) image that
displays the lines in green and the original image
in red.  Output is usually sent to xv.  I am adding
a variety of options, including thresholding by 
confidence, thresholding by absolute maximum number
of lines, theta range reduction, and r range reduction.
(Actually, all of these are done as of today).

The program is written in C++, but is understandable
to any C programmer.  I use no objects and the only
reason that it's in C++ is because I really find
the iostream.h library far superior to stdio.h.

				--Rich

PS:  These programs were written to analyze maps
created by an Histographic In Motion Mapping method
using the a sonar ring from a mobile robot.  If you're
interested I can also send that.

This was simply code that I had written, and I didn't spend 
a ton of time making it generalized and optimized.  In the
future, I have tentative plans to create a C++ Object Heirarchy
with pnm formated images.  Hopefully, if I complete that,
I can post it also.

------------------------------

Date: 	Thu, 5 Aug 1993 16:34:51 -0400
From: garym@vered.rose.utoronto.ca (Gary Lawrence Murphy)
Subject: IMSL routines in the Tsai article

This may be this month's most popular question ;-)

The camera calibration code for Roger Tsai's article "A Veratile Camera
Calibration Technique..." (IEEE Journal of Robotics & Automation 1987)
mentions the possibility of public domain replacements for the IMSL
statistics functions (dunlsf and dumiah)

These functions are described as "a nonlinear least squares using a
finite difference Jacobian and a multivariate function using a
modified newton method and a user-supplied Hessian"

The Question of the Week is: Does PD code or an algorithm for
functions like these already exist somewhere?

Gary Lawrence Murphy ---------------- garym@virtual.rose.utoronto.ca
University of Toronto, Industrial Eng Dept    fax:    (416) 978-3453
4 Taddle Creek Rd, Toronto, Ont M5S 1A4       voice:  (416) 978-3776
The true destination is always just here, now ----------------------

------------------------------

Date: Thu, 5 Aug 1993 19:17:27 -0500
From: cpg@research.att.com (Carlos Puchol)
Subject: A frame grabber device driver for Linux

I have written a frame grabber device driver for Linux (a free
version of UNIX). Although there is not scientific contents,
I think this is a very cheap alternative for those students around the
world willing to do their own vision applications without spending a
whole lot of money. We use the system for robot vision, telepresence
and remote sensing applications.

The device driver can be adapted to other versions of UNIX with
minor modifications. The board itself costs around $155 (US$).

Thank you.

++ Carlos Puchol

*****

This is a Linux device driver for the  WinVision frame grabber. The WinVision
video capture board is made by Quanta Corporation, and currently
costs $155, including PhotoFinish 1.0 by ZSoft for Windows. I have
nothing to do with them other than having bought one of their
video capture boards. At the end of the message you will find their info.

The current release as of August 5 is 1.0.

The frame grabber captures images in black and white and has a resolution
that depends on the incoming image, the format (PAL or NTSC) and some other
minor factors. My programs read images of 310x241 in NTSC format with
good aspect ratio and 187x241 in raw format (the image is expanded
to a good aspect ratio internally in the device driver if an IOCTL
is issued. See grab_exp.c for details). Some applications (like
vision & robotics, which is what I do), don't really need a good aspect
ratio for processing and can benefit better of some extra speed.

The depth of images captured is 6 bits per pixel, more than enough
for black and white images.

I shipped a few programs with this device driver for people to try:

grab_exp.c:	Grab images and display them in X. Push Q to quit,
		C for continuous mode and any other key to stop
		continuous mode.

grab.c:		Game as before, but with raw data, no correction for
		aspect ratio.

fgtopbm.c:	Output a pbm format greyscale image in stdout.

rate.c:		Determine the rate of capture of your board/machine.

Note: the continuous capture mode doesn't capture all the
frames coming out of the image source, but only the ones that are
available when the software is ready for the next capture. The image
displayed is the one coming out of the source at that precise moment
(i.e. the sound, if any is associated with the image, like in a VCR,
IS in sync with the image). The images have a "stroboscopic effect"
since the capture rate seldom goes above 5 or 6 frames per second.

I have tuned the device driver for NTSC images, but I think Quanta will
ship anywhere, and they have PAL versions. If you try to run the device
driver and associated programs with a PAL image, then remove the
#define NTSC_VIDEO line in the driver. You will most likely have to do
some adjustments in the variables that control the loops (the ones inside
the #define'd part). You can write a small program to do this using the
ioctl() calls. When you find a good set of parameters, please send them
over for use by some other people.

Read the file INSTALL for installation instructions.

Have Phun!

++ Carlos Puchol
++ cpg@cs.utexas.edu

The board company's info:

	Quanta Corp.
	2629 Terminal Blvd.
	Mountain View, CA 94043
	(800) 682 17 38
	(415) 967 57 91
	(415) 967 87 62 (FAX)

*****  Availability  

I have just uploaded fgrabber-1.0.tgz to the following locations:

        tsx-11.mit.edu:/incoming
        sunsite.unc.edu:/pub/Linux/Incoming

they will be there for a while until the system administrators decide
where to put it in the Linux ftp tree. You may also find them soon in
your favourite linux ftp mirror site. It is 9042 bytes long.

------------------------------

Date: Fri, 6 Aug 1993 12:47:46 GMT
From: ttj10@eng.cam.ac.uk (Tim Jervis)
Organization: Engineering Department, Cambridge University, England.
Subject: 3D object recognition using affine invariants

The following technical report is available by anonymous ftp from the
archive of the Speech, Vision and Robotics Group at the Cambridge
University Engineering Department.

                      Towards 3D Object Model 
                    Acquisition and Recognition 
                     using 3D Affine Invariants

                  Sven Vinther and Roberto Cipolla

                Technical Report CUED/F-INFENG/TR136

            Cambridge University Engineering Department 
                        Trumpington Street 
                        Cambridge CB2 1PZ 
                             England 


                             Abstract

We evaluate the power of 3D affine invariants in an object recognition
scheme. These invariants are actively calculated by the real-time
tracking of 2D image features (corners) over an image sequence. This
is done optimally by using a Kalman filter. Object information is
located in a hash table where it is stored and retrieved using the
invariants as stable indices. Recognition takes place when significant
evidence for a particular shape has been found from the
table. Preliminary results with real data are presented, and some of
the noise problems arising due to the weak perspective approximation
and corner localisation errors are discussed.

************************ How to obtain a copy ************************

a) Via FTP:

unix> ftp svr-ftp.eng.cam.ac.uk
Name: anonymous
Password: (type your email address)
ftp> cd reports
ftp> binary
ftp> get vinther_tr136.ps.Z
ftp> quit
unix> uncompress vinther_tr136.ps.Z
unix> lpr vinther_tr136.ps (or however you print PostScript)

b) Via postal mail:

Request a hardcopy from

Sven Vinther,
Cambridge University Engineering Department, 
Trumpington Street, 
Cambridge CB2 1PZ,
England.

or email me: sv@eng.cam.ac.uk


Posted on behalf of Sven Vinther by Tim Jervis, while Sven is
temporarily unable to access a terminal. Please address all
correspondence to Sven Vinther, sv@eng.cam.ac.uk.

--
 _/_/_/_/_/  _/  _/      _/   Cambridge University
    _/	    _/  _/_/  _/_/    Engineering Department,
   _/ 	   _/  _/  _/  _/     Trumpington Street,
      	                      Cambridge CB2 1PZ, England
--
 _/_/_/_/_/  _/  _/      _/   Cambridge University
    _/	    _/  _/_/  _/_/    Engineering Department,
   _/ 	   _/  _/  _/  _/     Trumpington Street,
      	                      Cambridge CB2 1PZ, England


------------------------------

Date: Thu, 05 Aug 93 20:44:33 -0400
From: David Coombs <coombs@cme.nist.gov>
Subject: CFP: Toward Physical Interaction and Manipulation

------------------------------------------------------------------------
------------------------------------------------------------------------
----------						      ----------
----------		     PRELIMINARY		      ----------
----------						      ----------
----------		CALL FOR PARTICIPATION		      ----------
----------						      ----------
------------------------------------------------------------------------
------------------------------------------------------------------------
----------						      ----------
----------						      ----------
----------  "Toward Physical Interaction and Manipulation"    ----------
----------						      ----------
----------						      ----------
----------	     AAAI SPRING SYMPOSIUM SERIES	      ----------
----------						      ----------
----------		 STANFORD CALIFORNIA  		      ----------
----------						      ----------
----------		  MARCH 21-23, 1994		      ----------
----------						      ----------
----------						      ----------
------------------------------------------------------------------------
------------------------------------------------------------------------

We are delighted to invite contributions for the 1994 AAAI Spring 
symposium: "Toward physical interaction and manipulation" to be held
on the campus of Stanford University, March 21-23, 1994.


SYMPOSIUM DESCRIPTION:

The range and scope of practical robotics applications depends
critically on the ability of robots to physically interact with their
environments.  Current applications are highly specialized, and
typically they involve carefully controlled, well understood
workspaces with little or no sensory feedback.  Construction costs and
inflexibility limit the economic viability of these systems. The
general manipulation skills of humans and other animals contrasts
starkly with the current capabilities of robots.  From threading a
needle, to opening a door, to catching a ball, to moving a sofa, we
engage our environments in myriad ways.  Unlike most current robots,
we rely upon rich sources of sensory feedback to cope with
uncertainties in our varied world.

The purpose of this workshop is to draw together researchers from a
range of disciplines to study the principles of physical interaction
and manipulation.  The goal is to consider theories, paradigms, and
ontologies for both natural and artificial systems, and to develop
generally useful concepts, architectures, and algorithms for building
and describing them.

The approach is to select in advance a set of tasks that range in
difficulty and span a number of research issues.  Each prospective
participant is to develop conceptual designs for one or more of these
tasks prior to the workshop.  It is acceptable for designs to be
speculative, as we encourage creative solutions.  However, the aim is
to examine tasks in detail and sketch complete systems. At the
workshop, selected designs will be presented, discussed, and compared
in an attempt to reach a more general understanding.  By analyzing a
range of tasks, we aim to broaden our perspective, identifying common
themes and useful design principles.  The rationale for this format is
that participants will be well prepared for the discussions by
thinking in detail about some of these tasks in advance.  The list of
candidate tasks follows:

  - make a cup of coffee
  - fry and serve an egg
  - prepare buttered toast
  - play catch
  - insert and play a video tape
  - vacuum/mop the floor or mow the lawn
  - dig a hole/trench
  - (un)lock a door with a key
  - open, pass through, and close a door
  - feed someone using a fork, knife, spoon, cup, etc.
  - retrieve a screwdriver from the toolbox in the garage.
  - fold clothes
  - move large objects (boxes, chairs, furniture)

These activities involve a range of skills and will most likely
require a range of mechanisms.  They can be characterized by their
requirements for:

  - real-time dynamics
  - ballistic vs. servo control
  - timed control
  - position/orientation/velocity/force control
  - tool usage & action at a distance
  - multiple temporal phases
  - sensor modalities (e.g., visual, haptic)
  - compliance
  - constraints on the workspace/environment

Participants should attempt to characterize their tasks and designs
according to these (and other) features to facilitate comparison.

SUBMISSION & PREPARATION: Potential participants should submit a short
description of their background and research interests along with
designs and analyses for individual tasks. To improve the depth and
quality of the designs, participants are encouraged to work in teams,
especially in collaborations that combine complementary expertise.  Of
course, demonstrations of working systems, including simulations and
videos, are encouraged.  Send submissions to either:
 
	Steven Whitehead                                              
	GTE Laboratories Incorporated           swhitehead@gte.com    
	40 Sylvan Rd.                           phone:  (617) 466-2193
	Waltham, MA 02254                       FAX:    (617) 890-9320

				  or

	David Coombs
	Natl Inst of Stds and Tech (NIST)	coombs@cme.nist.gov
	Robot Systems Division		
	Building 220, Room B-124		phone:	(301) 975-2865
	Gaithersburg, MD 20899 USA		FAX:	(301) 990-9688

ORGANIZING COMMITTEE: 
Emilio Bizzi, MIT; Jon Connell, IBM Watson; David Coombs, NIST,
co-chair, (coombs@cme.nist.gov); Ken Goldberg, USC; Rod Grupen, UMass;
Stan Rosenschein, Teleos Research; Steven Whitehead, GTE Labs,
co-chair, (swhitehead@gte.com);  

IMPORTANT DATES:

 	Submissions due:		 	October 15, 1993
	Notification of acceptance:		November 15, 1993
        Final registration deadline:		March 1, 1994
	Spring symposium:			March 21-23, 1994


------------------------------

Date: Tue,  3 Aug 93 17:54:17 +0300
From:  Anatoly L.Shishonok <leming@energetic.kiev.ua>
Organization: Energetic Coop.
Subject: Ukrainian work

                           The main aim

     To raise the work at artificial intelligence  in  Ukraine  to
the modern international level. To realize the idea it's important
to decide the following problems in  the  competitive  basis  with
suth financial support:
        - to  develop  theoretical  fundamental  research  in  the
          sphere of   computer   programming,   leading   to   the
          appearance of the new qualitative  properties,  increase
          of quick  effect  and  reliability  that  can decide the
          tasks of artificial intelligence;
        - development  of  mathematical  foundation of fuzzy sets,
          interval arithmetic, situational control etc, containing
          branches of mathematical theory of artificial control;
        - development  of  highly  productive  systems  using  the
          principles of artificial intelligence,  perfect means in
          personal contact with the machine in order to decide the
          most dofficult  scientific tasks in economic management,
          in creating the basis of knowledge;
        - development  of  new technologies of creating superlarge
          scale integration,   transducers   and    devices    for
          artificial intelligence systems;
        - preparation   of   scientific   specialists   concerning
          problems of artificial intelligence.
     The main tasks following from the main  aim  can  be  defined
more precisely.  They consist of the basis of computer engineering
with the use of the principles of artificial intelligence:
        1 project   of   the   elements   with  qualitatively  new
          properties;
        2 development  of  automatic  machines theory of the first
          and second type to automatic hierarchy theory;
        3 elaboration of calculative structures of non-traditional
          type having   parallelism   and   information   pipeline
          processing;
        4 elaboration of the basic processing parallel  theory  of
          general and   partial   information  leading  to  reduce
          overlapping variants in search dacisions:
        - artificial  intelligence  mayhematical  basic  extention
          tasks with new properties of  elements,  facilities  and
          computer engineering systems;
        - highly productive system  with  the  use  of  artificial
          intelligence principles;
        - tasks of creation and implantation of  new  technologies
          development facilities   for   artificial   intelligence
          system:
          a) super-large scale integration;
          b) transdusers;
          c) specialized devices.
     Tasks of preparation of scientific  personnel  in  artificial
intelligence problems;
        1 in Universities;
        2 in polytechnical institutes;
        3 in economic institutes (universities);
        4 in the academies of sciences;
        5 exchange of specialists with other countries;
        6 conducting of conferences, seminars;
        7 creation and issue of specialized journals.
     Financing of  these  works in the way of the help from abroad
will help  to  fulfil  the  scientific  research   of   artificial
intelligence problems  in  Ukraine because of lack of money.  Such
scientific staffs like the Institute of Cybernatics of the Academy
of Sciences, Calculative Control Machines Plant, Kvazar and others
are bankrupt and reduced.
     The loss  of  scientific sckools and research specialists and
stuffs woll throw back for decades the development of Ukraine  and
hamper its integration in the world's community.
     One of the branches of elaboration of computer programming of
new architecture  that  are  built  on the artificial intelligence
principles and designed for  creation  of  effective  intellectual
systems in the internal intellectualization of computers.

     Commerce offer  for mutual collaboration in the sphere of new
                  computer engineering.

     1. The research substance.
     L.F.Marakhovsky was   the  first  to  create  multifunctional
structures and multilevel schemes of memory (MS and  MSM)  on  the
potential and dynamic types of elements of large scale integration.
Comparison of  created  MS  and  MSM  with  known   monofunctional
multistable triggers  -  K-external elements "AND-NO" with pull-up
capability P1  and  R-external   elements   "AND"   with   pull-up
capability P2 (K = 8;  R = 8;  P1 = P2 = 10) are given in tables 1
and 2.

     Table 1. Parametres of basic memory schemes.
*
             Monofunctional   MS       MS   Preferrable     
 Parameters  multistable                  m memory scheme   
             triggers (MMT)  cl. L   cl. L  class           
4
                                                      m     
 M max              8          30      90     MS cl. L      
4
                                                            
 F w             1/4 N e    1/4 N e 1/6 N e   MS cl. L      
4
                                                      m     
 N q                3           6       9     MS cl. L      
4
                                                      m     
 S con       if limited ele-   18      12     MS cl. L      
 with M-28   ments permit                                   
4
                                                            
 W                  1          0.6     0.7    MS cl. L      
4
                                                           m
 N e                         more 8  more 8  MS cl. L and L 
)

 M max  -    maximum number of storage  status  with  limited  lo-
             gical elements
 F w    -    work  frequency  (maximum) with N e (one logical ele-
             ment delay)
 N q    -    pull-up capability memory
 S con  -    number  of  interelement  connections  in  the memory
             scheme
 W      -    equipment consume for one memory scheme status
 N e    -    the namber of different blocks of memory storage with
             definite (optional  external)  signal,  allowing   to
             access alternative   mapping   without   delay  which
             doesn't exist in triggers.

     Table 2.  Used logical elements  quantity  and  their  limits
               while construction  basic memory schemes storing 18
               conditions.
*
 Logical                                     Dominant       
 elements          MMT     MSM      MSM  k   memory scheme  
 parameters               cl. L n  cl. L n   class          
4
 Number of                                             k    
 elements          18       14        12     MSM cl. L n    
                                                            
4
 Number of                                             k    
 exits of          18        4         3     MSM cl. L n    
 elements                                                   
4
 Number of                                             k    
 pull-up           18        5         4     MSM cl. L n    
 elements                                                   
)

     We can  see  the  advantages  of  allowed memory schemes from
these  two   tables.  External  parameter   N e   reflects   their
quantitively new properies.  Monofunctional trigger fasilities are
their partial case.
     Creation of  qualitatively  new  memory  schemes enlarged the
basis of computer engineering that allowed to  widen  the  machine
theory of  the  1st  and 2nd type and to create the machine of the
3rd type with qualitatively new big transition.  We can create the
hierarchic machine theory, allowing to perform partial information
processing parallelly with general information  on  the  basis  of
machine theory  of the 3rd type.  Of allowed to increase the speed
of interconnection between their levels with N  (N  level)  clocks
till the delay of the first logical element.
     While creation of N (N > 3) level machines we can fulfill not
only determinated  transition as we did it for the machines of the
first and second type (from environment  to  environment)  and  to
realize the digital transition from one set environment (block) to
another defined  subsets  environment  and  also  to  fulfill  the
probable transition  to fuzzy subset,  consisting of quite defined
environment elements. That principally couldn't be realized before
in trigger devices.
     From economic point of view  a  newly  appeared  direction  -
elaboration of computer with qualitatively new memory organization
multifunctional system  allowed  to  create  the   computer   with
qualitatively new  probabilities  with high speed and influence of
the general information on processing of partial one.
     So we  elaborated  the  original  discrete  devices of design
theory with multifunctional memory organization system on  MS  and
MSM.

     2. The substance of commercial proposition.

     2.1 Together   with   computer  designers  to  elaborate  the
conception of a new computer  with  qualitatively  new  properties
using the elaborated device design theory on MS and MSM.
     2.2 To design the Technical job for software and hardware.
     2.3 To design the system and application software.
     2.4 Design of technical computer project:
         - structure (architecture)
         - super-large scale integration (SLSI)
         - constructions and printed circuit
         - the  design  of  diagnostic  hardware  and   diagnostic
           software.
     2.5 Production of super-large  scale  integration,  circuits,
constructions, test hardware and the computer itself (experimental
sample).
     2.6 Debugging   of   systematic   applicative  and  technical
software.
     2.7 Preparation of documentation for experimental sample.

     The author of the commercial proposition could take an active
part in p.2.1;  2.2; 2.4; 2.6 and 2.7 and also lead the scientists
taking part in p.2.4.

                                                   L.F.Marakhovsky

     252028, kIEW-28
     PROSPEKT nAUKI 35 KORPUS 4 KW. 23
     l.f.mARAHOWSKIJ
     T.(044) 265-77-39

------------------------------

End of VISION-LIST digest 12.35
************************
