The replies you received from BIOMECH-L readers citing various
ambiguities of the orientation or position of "sticks" with respect to
the image plane of a single camera left out the simple observation that
the human body is not a stick. While self occlusions of the human body
are annoying, in many cases they are not fatal computationally as you
can fill in the data holes. Clinical gait analysts may reject data
filling or extrapolation methods out of hand but you may have a
different audience in mind.

I suggest you have a look at methods which use additional data to fit
the projection of the human body pose and movement to a 3D pose. This
is typically called "markerless video based motion capture". Here are
some references from a couple of years ago, I'm sure the field has
advanced since then.

Aggarwal, J.K. and Q. Cai (1997). Human Motion Analysis: A Review. In
Proceedings of the IEEE Nonrigid and Articulated Motion Workshop, pp.
90-102. June 15-16, San Juan, Puerto Rico.

Cedras, Claudette and Mubark Shah (1994). A Survey of Motion Analysis
from Moving Light Displays. In CVPR94, pp. 214-221.

Deutscher, Jonathan, Andrew Blake, and Ian Reid (2000). Articulated
Body Motion Capture by Annealed Particle Filtering. In Proceedings

Leventon, Michael E. and William T. Freeman (1998). Bayesian estimation
of 3-d human motion from an image sequence. Technical Report 98-06,
MERL - A Mitsubishi Electric Research Laboratory.


Jason Harrison, PhD
Post-Doctoral Fellow and Imager Lab Manager
Imager Laboratory for Graphics, Visualization and HCI
FSC3640 - 2424 Main Mall, Vancouver BC V6T 1Z4
Cell: 604 644 8611 Lab: 604 822 2218; Fax: 604 822 8989

To unsubscribe send SIGNOFF BIOMCH-L to
For information and archives:
Please consider posting your message to the Biomch-L Web-based
Discussion Forum: