Tweet
Date: Wed, 17 Jun 92 17:05:37 -0400
From: ngoddard@carrot.psc.edu
Subject: Recognizing human gait: Dissertation & data available via ftp
I have placed the human motion sequences used in my thesis in the file
pub/outgoing/public-mld-data.tar.Z on ftp.psc.edu (128.182.62.148).
This file contains the raw data, a smoothed version of some of it, and
code for extracting and displaying the motion sequences. It is
contributed to the community with no guarantee of follow-up. I ask
only that you make appropriate acknowledgement in publications which
use the sequences or code, and notify me (ngoddard@psc.edu) of the
publication. I expect they could be of use for people working on
stereo, correspondence, tracking, structure-from-motion, and
recognition. My dissertation:
"Perception of Articulated Motion: Recognizing Moving Light Displays"
is available for ftp from cs.rochester.edu (192.5.53.209) in file
pub/papers/ai/92.tr405.moving_light_displays.ps.Z. It is more fun if
printed double sided. The short story is: actor-independent gait
recognition is achieved using a novel parallel attention mechanism,
requiring about one second of input (people need about half a second).
These motion sequences are taken from a WATSMART gait analysis system.
Each frame contains the location of 8 markers which where attached to
the joints of a person who was moving (walking, running, etc) roughly
parallel to the image plane. The marked joints were distal (far)
wrist and ankle, and proximal (near) shoulder, elbow, wrist, hip, knee
and ankle. There are raw 2D data from a pair of cameras (giving a
stereo pair) and 3D data computed by the WATSMART software from the 2D
data. The 3D locations are given in tenths of a millimetre. Since
motion was roughly parallel to the image plane, there is almost no
motion in depth. The frame rate is 100 frames/second. There are
several samples of walking, running, skipping, running on the spot and
possibly other movements from four individuals, two male and two
female. The smoothed data provides several samples of complete cycles
of walking, running and skipping from each of the four actors.
Nigel Goddard
ngoddard@psc.edu
From: ngoddard@carrot.psc.edu
Subject: Recognizing human gait: Dissertation & data available via ftp
I have placed the human motion sequences used in my thesis in the file
pub/outgoing/public-mld-data.tar.Z on ftp.psc.edu (128.182.62.148).
This file contains the raw data, a smoothed version of some of it, and
code for extracting and displaying the motion sequences. It is
contributed to the community with no guarantee of follow-up. I ask
only that you make appropriate acknowledgement in publications which
use the sequences or code, and notify me (ngoddard@psc.edu) of the
publication. I expect they could be of use for people working on
stereo, correspondence, tracking, structure-from-motion, and
recognition. My dissertation:
"Perception of Articulated Motion: Recognizing Moving Light Displays"
is available for ftp from cs.rochester.edu (192.5.53.209) in file
pub/papers/ai/92.tr405.moving_light_displays.ps.Z. It is more fun if
printed double sided. The short story is: actor-independent gait
recognition is achieved using a novel parallel attention mechanism,
requiring about one second of input (people need about half a second).
These motion sequences are taken from a WATSMART gait analysis system.
Each frame contains the location of 8 markers which where attached to
the joints of a person who was moving (walking, running, etc) roughly
parallel to the image plane. The marked joints were distal (far)
wrist and ankle, and proximal (near) shoulder, elbow, wrist, hip, knee
and ankle. There are raw 2D data from a pair of cameras (giving a
stereo pair) and 3D data computed by the WATSMART software from the 2D
data. The 3D locations are given in tenths of a millimetre. Since
motion was roughly parallel to the image plane, there is almost no
motion in depth. The frame rate is 100 frames/second. There are
several samples of walking, running, skipping, running on the spot and
possibly other movements from four individuals, two male and two
female. The smoothed data provides several samples of complete cycles
of walking, running and skipping from each of the four actors.
Nigel Goddard
ngoddard@psc.edu