H.264 motion estimator design

Official URL: http://risc01.sabanciuniv.edu/record=b1147594 (Table of Contents)

Recently, a new international standard for video compression named H.264 / MPEG-4 Part 10 is developed. This new standard offers significantly better video compression efficiency than previous international standards. The variable block size motion estimation is the most compute-intensive part of an H.264 video encoder. The full search method is impractical for real-time implementations since it requires a high computational complexity. Therefore, many fast motion estimation algorithms have been developed for real-time implementations. In this thesis, we used an SAD reuse based hierarchical motion estimation algorithm for real-time H.264 / MPEG-4 Part 10 video coding. This algorithm uses the Lagrangian cost parameter (SAD+λR) for selecting the best motion vector. We designed a high performance and low cost hardware architecture for real-time implementation of this algorithm. We have considered several alternative designs and decided on this architecture based on a cost/performance analysis. This architecture uses a novel data flow resulting in a low cost and high performance hardware. This hardware is designed to be used as part of a complete H.264 video coding system for portable applications. The proposed architecture is implemented in Verilog HDL. The Verilog RTL code is verified to work at 63 MHz in a Xilinx Virtex II FPGA. The FPGA implementation can process 25 VGA frames (640x480) or 76 CIF frames (352x288) per second.