An efficient motion vector prediction method for avoiding AMVP data dependency for HEVC

Abstract

To achieve higher coding efficiency, the latest video coding standard called High Efficiency Video Coding (HEVC) has adopted the mechanism of Advanced Motion Vector Prediction (AMVP) to further improve the accuracy of motion vector predictor. However, the adoption of AMVP significantly increases the hardware realization overhead as well as the data access bandwidth requirements. In addition, the dependency between different coding units (CUs) or prediction units (PUs) for predicting AMVP also noticeably degrades the overall hardware coding throughput. To deal with this problem, this paper proposes an efficient motion vector prediction method for avoiding AMVP data dependency. By modeling the relationship between motion vector predictors of largest coding unit (LCU) and other small CU and PU sizes, the motion vectors of small CUs and PUs are estimated directly from the motion vectors of LCU. Furthermore, the predicted motion vectors of small CUs and PUs are also used to pre-fetch the corresponding reference data from external memory in advanced so that the data access time can be hided. Simulation results demonstrate that the proposed motion vector prediction method can achieve at least 53.8% coding throughput improvement with only 1.04% BD-rate increasing when compared to direct AMVP realization.