in various applications. Capturing of human motion
in real-time and with high accuracy is currently achieved by constraining
the capture environment, e.g. with markers on the person, skin thight clothes,
homogenous background and lighting etc. Tracking and capturing the motion of
a human with arbitrary clothes before an arbitrary background and varying
lighting conditions is a very challenging problem. Presented in this talk are
approaches towards the solution of this difficult and complex task.
In the first part of the talk an interaction
system is presented that is capable of tracking the head and hand of a person
in real-time. The user is acting within a 3-sided CAVE that utilizes openSG
for multidisplay projection.
The second part concentrates on markerless optical motion capture.
The movement capabilities of
the human body are modeled as an articulated object, which constrains the
possible movements and allows Nonlinear Least Squares methods for estimating
the current pose.
To achieve correct results with cluttered unknown background
and more general clothing, the motion estimation incorporates
different features like depth information from stereo,
gradient/histogram based sillhoutte methods and KLT image features.