[comp.robotics] creating 3D motion files of bipeds walking, etc

karl@ficc.ferranti.com (Karl Lehenbauer) (06/27/90)

What is a good way to go about creating motion files of people walking,
running, kneeling, sitting down, etc, etc?  I would like to record positions 
of hands, elbows, shoulders, waist, knees, ankles, etc, over time.

Are there some kind of really small intertial units that would be accurate
to .2" or so for a few seconds within a 100 sq. ft. area that wouldn't be
cost-prohibitive?

What else?  I thought of using bright dots a la the brute force Robert Abel 
and Associates approach for the commercial of the female robot for canned
corn, but the result's 2-D and require a lot more hand-work to use.
-- 
-- uunet!ficc!karl  	"Fun has a name, and it's Bartholomew J. Simpson"
   uunet!sugar!karl

arh@sage.cc.purdue.edu (Eric B) (06/28/90)

In article <1N945P5@xds2.ferranti.com> karl@ficc.ferranti.com (Karl Lehenbauer) writes:

>What is a good way to go about creating motion files of people walking,
>running, kneeling, sitting down, etc, etc?  I would like to record positions 
>of hands, elbows, shoulders, waist, knees, ankles, etc, over time.

   I saw a report on CNN-Headline News about a group doing this.  Their
focus was "How much time does a person spend in any position during a 
typical workday?"
   The report showed a man wired with "mercury switches" (about 3cm square,
1cm thick) that sensed orientation in 2D.  The info from the switches was
stored in metal box that the person carried in his pocket.  At the end of
the day, the box was plugged into a computer which generated a histogram
of the time spent per position.  
   They also showed a real time representation of position (motion) with
the man connected to a pegboard with sets of lights for each position 
like so...

      body			arm

	|			|\
	|   /   		| \
	|  /   /		|  \
	| /  /			|_ _\_ _ _
	|/ /
	|/_ _ _			  /
	      |\		 /---
	      | \_		/ \
	      |_

   I don't intend for this to be a complete description of what the group 
is doing, only what Headline News conveyed to me.
   This may not be what you want, but if you talk to your local cable
company, they should be able to help you get in contact with CNN.

						later,

						Eric G. Bolinger  8-)
						arh@sage.cc.purdue.edu

nagle@well.sf.ca.us (John Nagle) (06/29/90)

      There are several basic approaches.  How much money have you got?

      First, there's the VPL DataSuit (tm).  This contraption is an 
instrumented wetsuit, with angle measurement at all major body joints
and position measurement for the back and head.  Not useful for large
movements; the umbilical tether is big, heavy, and fragile.  Contact
VPL, Redwood City, CA.

      Direct movement generation has been done.  See the SIGGRAPH
videos.  Girard's "Eurythmy" has some good human animated humans.
The paper is in the 1988 SIGGRAPH proceedings.

      I like Mike Kass's approach, but he hasn't developed it to the
point that human movement emerges.  I'd suggest work in this area as a
PhD thesis topic for someone.

      Various body recording schemes have been developed.  If you wanted
to record body movement without interfering with motion, I'd suggest putting
your subject in a unitard with an irregular pattern of angled lines and dots,
videotaping from multiple angles, aligning the images via convolution, and
extracting the limb positions.  Non-trivial.

						John Nagle