Real-time Human body tracking for public spaces

Kushal Vaghani

Dept of CSSE, University of Canterbury

Fri Oct 01 15:30:00 NZST 2004 in Room 031, MSCS

Abstract

We develop a novel algorithmic technique that can be used to estimate 3D joint positions for a human body in a public space. This information can be used for motion capture and other vision-based applications. Previous motion capture applications either rely on joint markers or are not robust enough to work in public spaces. The problem to solve in a motion capture application is robust tracking of human body parts. We develop and test a technique which can robustly detect and track joint positions of upper human body in an unconstrained environment. Our results show that the estimates of the 3D head, shoulder and torso positions from our technique are sufficiently robust to the changing environment in a public space.