A Hardware-Efficient VLSI Architecture for Variable Block Size Motion Estimation in H.264/AVC

Wenwen Zhang,Yong Ding,Xiaohua Luo,Gang-De Sun
DOI: https://doi.org/10.1109/ICECC.2012.11
2012-01-01
Abstract:A hardware-efficient VLSI architecture for full search variable block size motion estimation with is presented in this paper. By combining snake search strategy with intracandidate 2-parallel data reuse strategy, this design can save 45%~60% local memory accesses than that snake search strategy and that raster search strategy. It can get 41 motion vectors of one MB in parallel with data reuse of the SADs of smaller size blocks. Also, our design can simultaneously compute two candidates in parallel in one cycle. Synthesized with SIMC 0.18µm standard cell library, the proposed design takes 180K logic gates and 3.15KB local memory. For the search range of [-16~+15], it allows the real-time variable block size motion estimation of 1920脳1088@30fps at 140MHz.
What problem does this paper attempt to address?