Technical Report TR2006-08-01 August, 2006
Internetworking and Media Communications Research Laboratories
Department of Computer Science, Kent State University
http://medianet.kent.edu/technicalreports.html


 

Thesis Title

AN ALGORITHM FOR FAST PERCEPTUAL OBJECT TRACKING IN A CODED STREAM BY ANALYSIS OF COMPOSITE SCENE MOTION’S REFLECTION ON THE MOTION VECTORS

Zhong Guo
mail to: zguo1@kent.edu

Supervisor: Prof. Javed I. Khan
Department of Computer Science
Kent State University

Date Submitted: August 2006


Abstract

In this research we developed an object tracking algorithm suitable for perceptual encoding. Object based bit allocation can result in significant improvement in the perceptual quality of relatively low bit-rate video. However, a particularly computationally challenging aspect in the process is the video object tracking. This is because traditional algorithms are based on classical object detection techniques. In this thesis we present an algorithm particularly designed for perceptual video coding. The algorithm is optimized for minimum decoding and uses only motion information and detail case analysis that accounts for variety of cases that arise in motion vector coding due to relative scene and camera movements. The result is a fast yet highly effective perceptual region tracking algorithm that can operate in stream rate and track regions of perceptually significant object despite camera movements such as zoom, panning and translation. We have implemented this algorithm into a live MPEG-2 perceptual transcoder. We also share the performance of its MPEG-2 compliant real implementation. This fast object aware video rate transcoder is particularly suitable for live streaming and can convert a regular coded video into an object-based perceptually coded video stream.



Last Modified: August 2006.