Adaptive Model-Based Multi-person Tracking

Abstract

This paper proposes a method for tracking and identifying persons from video image frames taken by a fixed camera. The majority of conventional video tracking surveillance systems assumes a likeness to a person's appearance for some time, and existing human tracking systems usually consider short-term situations. To address this situation, we use an adaptive background and human body model updated statistically frame-by-frame to correctly construct a person with body parts. The formed person is labeled and recorded in a person's list, which stores the individual's human body model details. Such recorded information can be used to identify tracked persons. The results of this experiment are demonstrated in several indoor situations. © Springer-Verlag 2004.