A hardware friendly fractional-pixel motion estimation algorithm based on adaptive weighted model

In this paper, a new interpolation-free algorithm for fractional pixel motion estimation (FME) is proposed based on the statistical analysis of the matching error surface around the integer-pixel motion vector location. Performance analysis shows that 96% saving can be achieved for FME computation cost at the expense of only 2.2% BD-rate increment. The proposed algorithm was prototyped using Verilog and synthesized targeting 65nm CMOS technology. The design achieves a maximum frequency of 602 MHz and a high throughput where it can process 71 fps for FQHD (3840×2160) videos. It has only 26.6 kgates and power consumption of 18.6 mW.