ICRA 2005 tutorial
Visual Tracking: 2D/3D Methods and Cue Integration
Ville Kyrki |
Centre for Autonomous Systems
Royal Institute of Technology
Lappeenranta University of Technology |
Department of Information Technology
Vision is one of the most powerful sensory modalities in the field of robotics. Longterm research in application areas such as service robotics, intelligent vehicles, medical robotics, surveillance, and human-robot interaction has resulted with different types of visual tracking algorithms. The aim of this tutorial is to present theoretical basics of visual tracking and overview some of the most common methods. We will spend time on issues such as the choice and estimation of visual cues and their integration for both 2D and 3D tracking.
* visual tracking basics
* 2D visual cues
* visual tracking in 3-D
* cue integration
* motion models
The advances in available computing power and the introduction of new methods in terms of sensors, control, and recognition have in recent years allowed design of a new generation of vision systems. Today, vision offers a flexible modality for generation of feedback to robot control systems in application areas such as service robotics, intelligent vehicles, medical robotics, surveillance, and human-robot interaction. Early research relied on relatively simple methods and visual cues in order to provide real-time performance. Recent methods require more computational power but the theoretical basis for many of the methods lies still in old ideas.
The objective of this tutorial is to provide a solid overview of
basic issues and ideas related to vision based tracking.
Even if different application domains offer different constraints and
impose specific requirements, many of the basic methods are the same.
In many cases, some of the methods used in the computer
vision community are not in general use in the robotics
community. There is therefore a need to present and discuss these in
terms of robotics research.