Robust motion estimation techniques

Abstract

[ACCESS RESTRICTED TO THE UNIVERSITY OF MISSOURI AT AUTHOR'S REQUEST.] Motion estimation is a very important step in many video processing tasks, including video compression, object tracking, and etc. A video may experience both global motion and local object motion. Global motion includes camera rotation, zooming, and changes in camera location and perspectives. In this thesis, we study different robust motion estimation techniques, such as block matching, motion intensity profile technique, and global motion estimation. Block matching has been widely used for motion estimation in video compression. This approach performs very well when there is only translational motion in the video sequence and fails in other types of camera motions, such as rotation and zooming. To address this issue, we propose a new motion search technique, called intensity profile. It characterizes a pixel using the intensity distribution in its neighborhood. Based on this intensity profile, we develop a distance metric for local motion estimation. This scheme can be further extended for global camera estimation. Our extensive experimental results demonstrate that the proposed motion estimation scheme based on intensity profile outperforms conventional block matching algorithm.