2004-2005 Academic Year

DESCRIBING AND UNDERSTANDING HUMAN BODY MOTIONS: THE LIGUISTIC 
POSTURES AND GESTURES
Speaker: ZORAN DURIC

Humans are articulated objects composed of non-rigid parts. We are 
interested in describing and recognizing their motions over various 
periods of time. Such motions can be understood with the aid of 
descriptive languages incorporating the hierarchical organization of 
the human body and the piecewise simplicity of motions of individual 
body parts. At the lowest level body motions are expressed in 
numerical terms, at higher levels they are usually expressed in 
symbolic terms. I will start by presenting some recent research 
results on detecting and tracking human movements. I will then 
demonstrate signal to symbol transformations to derive symbolic 
descriptions of human body motions on three examples. The first 
example will show how hand gestures can be used to replace a computer 
mouse. I will discuss the choice and recognition of gestures 
corresponding to various mouse actions. The second example will show 
how eye detection and tracking in color videos can be used for 
determining cognitive and affective states of a person. The third 
example will show the use of image flow for detecting and recognizing 
simple movements of body parts in color videos of humans.