Workshop on QoE-Aware Resource Allocation for Multimedia Communications

Research Article

A Novel Data Reuse Method For Motion Estimation In Video Applications

Download457 downloads
  • @INPROCEEDINGS{10.4108/eai.13-7-2017.2270544,
        author={Weizhi Xu and Hui Yu and Xin Wang and Yanhui Ding and Dianjie Lu and Guijuan Zhang and Zengzhen Shao},
        title={A Novel Data Reuse Method For Motion Estimation In Video Applications},
        proceedings={Workshop on QoE-Aware Resource Allocation for Multimedia Communications},
        publisher={EAI},
        proceedings_a={QOE-RAMC},
        year={2017},
        month={12},
        keywords={motion estimation memory traffic fruc},
        doi={10.4108/eai.13-7-2017.2270544}
    }
    
  • Weizhi Xu
    Hui Yu
    Xin Wang
    Yanhui Ding
    Dianjie Lu
    Guijuan Zhang
    Zengzhen Shao
    Year: 2017
    A Novel Data Reuse Method For Motion Estimation In Video Applications
    QOE-RAMC
    EAI
    DOI: 10.4108/eai.13-7-2017.2270544
Weizhi Xu1,*, Hui Yu2, Xin Wang2, Yanhui Ding1, Dianjie Lu1, Guijuan Zhang1, Zengzhen Shao1
  • 1: School of Information Science and Engineering, Shandong Normal University
  • 2: School of Management Science and Engineering, Shandong Normal University
*Contact email: xuweizhi@sdnu.edu.cn

Abstract

Motion estimation is a kernel algorithm in many video applications. Full search integer motion estimation (FSIME) can find the best result but it usually takes plenty of time. Previous work on accelerating FSIME exploited intra-frame data reuse within reference frame. We propose a novel data reuse scheme which uses not only intra-frame but also inter-frame data reuse for FSIME. Motion estimation in frame rate up-conversion (FRUC-ME) is used as a case study. A frame is loaded into on-chip memory only once instead of twice and used for two interpolated frames. Different inter-frame data reuse methods are presented and analysed. They give useful tradeoff between off-chip bandwidth requirement and on-chip memory size. The proposed data reuse methods all show better data reuse efficiency than the traditional methods, so the off-chip memory traffic is reduced effectively, as much as 37.5%.