Parallel motion estimation on the MDSP multiprocessor

  • Authors:

  • Yen-Chun Lin;Chuan-Da Hsiung

  • Affiliations:

  • Dept. of Computer Science and Information Engineering, National Taiwan University of Science and Technology, Taipei, Taiwan;Dept. of Computer Science and Information Engineering, National Taiwan University of Science and Technology, Taipei, Taiwan

  • Venue:
  • Neural, Parallel & Scientific Computations
  • Year:
  • 2007

Quantified Score

Hi-index 0.00

Visualization

Abstract

Motion estimation (ME) is a process for removing temporal redundancies in video sequences. It contributes most of the compression ratio and consumes typically 60-80% of the video encoding time. This paper investigates parallel execution of ME on the Multiprocessor DSP (MDSP) architecture, a software programmable, scalable multiprocessor platform. We use a low-overhead data distribution scheme and achieve load balancing among processors. In addition, we assign different tasks of an ME algorithm to suitable types of processors. Many sequential ME algorithms can be adapted to run on the MDSP in parallel, and the experimental results show that very good speedup can be obtained. Since the MDSP is software programmable, our design can be easily integrated into a video encoder, modified to meet different standards, and used in various video applications.