Technical Report Optical Ow Estimation 1 Optical Flow Estimation

The recent development in image processing and analysis, moving from static images analysis to dynamic image analysis, has focused the research on the understanding of motion analysis and representation. The psychological evidence has shown that human perception system is more sensitive to moving objects than static objects. We have the ability to recognize diierent objects and their motion in the real world scene. Therefore, the motion characteristic of objects and how to represent them have been studied by many researchers over many areas, e.g., digital image processing, digital video processing, and computer vision. In digital video processing, the data to be used for motion analysis and representation are mainly sequence of images or frames. This image sequence processing involves a large amount of data. Therefore, the important problems are how to process it in a reasonable amount of time, how to represent it in the compact form, and how to store it by using economical space. Even though the rapid progress both in computer technology and motion analysis and codec algorithms has contributed to these tasks being completed in a reasonable amount of time, to create more eecient and eeective algorithms is still an the important research topic. Before discussing in more detail, the two main deenitions of motion estimation , the 2-D motion eld and the optical ow, should be mentioned rst. The 2-D motion eld, or "projected motion", is the (perspective or the orthographic) projection of the 3D motion in real world scene on the 2-D image plane. The optical ow is the eld in image plane that is associated with the spatio-temporal variations of intensity pattern.