A Parallel Streaming Motion Estimation for Real-Time HD H.264 Encoding on Programmable Processors

Abstract

Motion estimation is an important computing intensive component in most video compression standards. The high computational costs and heavy memory bandwidth requirements of motion estimation give huge pressure to most existing programmable processors, especially in real-time high definition H.264 video encoding. Emerging stream processing model supported by most programmable processors provide a powerful mechanism to achieve high performance in media processing, which brings an opportunity to relieve this pressure. This paper presents a parallel motion estimation algorithm based on stream processing. Many approaches are explored to enable high data reuse efficiency and computation parallelism for GPUs or other programmable processors. Experiment results of 1080p H.264 encoding show that our parallel streaming motion estimation applied on different programmable processors accomplishes obvious speedup (from1.8x to 8.1x) over the serial motion estimation. Furthermore, the results show that H.264 encoder achieves 30FPS performance which satisfies the real-time requirements of 1080p H.264/AVC.