Tracking Multiple Objects
from Single, Multiple or Moving Uncalibrated Cameras
Abstract
We address three key elements of the video-based activity analysis in
various types of camera configuration. The three aspects are detection,
tracking, and multiple views integration. We present an approach to perform
robust visual-based multiple objects detection and tracking from a single or
multiple uncalibrated cameras for video-based activity understanding. It is
anticipated that this work will also provide and demonstrate a framework for
combining multiple sources of information from various sensors (e.g.
monocular information, 2D and 3D motion, and structures).