Motion Vector Prediction (MVP) plays an important role in improving coding efficiency in HEVC, H.264/AVC and AVS video coding standard. MVP is implemented by exploiting redundancy of adjacent-block optimal coding information under the constraint that MVP must be performed in a serial way. The constraint prevents parallel processing and MB pipeline based on LevelC+. In multi-stage pipeline, to some extent, adjacent-block best mode-decision information can hardly be obtained. In this paper, we propose a new hardware-oriented method to improve the coding performance at a cost of few hardware resources. When adjacent block is not available, spatial motion vector prediction (SMVP) for integer motion estimation (IME) and fraction motion estimation (FME) will take the IME best mode information and FME best mode information of left block as best information to derive PMV (Predicted Motion Vector) for current macro-block or block. Experimental results shows that the method we propose can achieve a better performance than the existing methods by 0.1db for the cases with intense movement and a non-degrading performance for flat cases.
Efficient motion vector prediction scheme for high speed motion estimation in H.264/AVC
