Demo Software for Performing Human Tracking
I have developed a Matlab program that estimates human motion in images taken with multiple
synchronized cameras using a simple body model based on
truncated cones. At Brown we have our own Motion Capture Lab
that we used to capture motion data for quantitative
evaluations.
Latest Version: Ver 1.0 (11/26/2005)
This product provides a basic Matlab implementation for
tracking a moving person in a sequence of multiple synchronized images, with
known camera calibration information and body dimensions. In
addition, evaluation code is provided to compare the tracking results to the
ground truth data.
Video Demo
In the video, the black outline represents the true pose
of the moving person. The colored outline represents the estimated pose by
this software.
Our implementation uses an annealed particle filtering
approach [2] with a few additions. For details
about our work and for a detailed set of experiments with this code see [1].
Results suggest that error accumulation seems to be under control as the
tracker recovers from errors.
For bug reports, questions, suggestions and comments,
feel free to
email
me.
[1]
A. Balan, L. Sigal and M. Black.
A
Quantitative Evaluation of Video-based 3D Person Tracking.
IEEE Workshop on VS-PETS, 349-356, 2005.
[2] J. Deutscher
and I. Reid.
Articulated Body Motion Capture by Stochastic Search.
IJCV, 61(2):185-205, 2004.
