A Novel Fast Motion Estimation Algorithm For H.264/AVC Based On Embedded System

With the development of video technology, video compression has been widely used into various fields, especially in the embedded system field, such as set-top boxer, videophones and so on. So the video compression and coding technology based on embedded system has significance not only for the research but also for the commercial products.In order to implement the embedded video system with low power, small size, high integration, we need to design different video systems for different chips. The performance of video compression will play an important role in the embedded video system.From the research of video coding and compression, Motion estimation (ME) is the main bottleneck and by far the most time-consuming module about 60%-80% computation in video coding application. Block matching method is widely used for video coding standards, such as H.263 and MPEG4. Motion estimation for H.264/AVC is different from them. Many new strategies with high complexity such as one-quarter pixel interpolation and multiple reference frames are presented. So some methods will be used to improve performance.Based on embedded system, a novel fast motion estimation algorithm for H.264/AVC is presented in this paper. Adaptive search algorithm consists of two stages: 1) stationary block search with zero-motion detection and based on motion vector predict stationary block search. 2) non-stationary block search with adaptive search and refined stage with diamond search. Furthermore we present fast variable-block-size search algorithm instead of variable-block-size partitions search in H.264/AVC.Finally we implement the proposed algorithm based on Blackfin ADSP-53X and use the optimization technology of Blackfin instruction set such as parallel processing architecture, video instruction and so on. Experimental results demonstrate that our proposed algorithm achieves good performance and can adapt to embedded video system.