CAD, computer vision systems for analysis of human activity

By
Larry Davis

                  Abstract

During the past decade, we have been studying problems related to
detection, tracking and behavioral analysis of humans in action.  This
talk will review that research, emphasizing our work on real time
vision systems for visual surveillance.  I will discuss the W4 system
- a real time PC based system hat can detect and track people and
their body parts, segment small groups of people into individuals and
recognize when people are carrying objects.  W4 uses a single
stationary camera so that it can employ background modeling techniques
for detection.  More recently, we have developed real time algorithms
for detection and tracking of people that can operate from a moving
platform.  These algorithms employ hierarchical shape models, coarse
to fine image matching using a variation of Hausdorff matching, and an
efficient sampling method to drive a condensation approach to
tracking.