Abstract:

Motion estimation is the most time consuming module with the highest computational complexity in the H.264/AVC system.In order to speed-up the encoding procedure,a parallel algorithm of H.264/AVC motion estimation is proposed in this paper based on a heterogeneous system composed of the computed unified device architecture(CUDA).A sequential search pattern is adopted in the algorithm,and the sum of absolute difference(SAD) information for other partitions is build with the SAD information for the smaller MB partitions.The algorithm is divided into three steps:obtaining SAD information for the smaller MB partitions,building SAD information for other partitions,and reduction to get the best result.The above three steps need to be executed sequentially but there are many threads executed parallel in each step.Experiment result shows that the execution time of the proposed motion estimation algorithm on NVIDIA's GTS450 platform is 50 times faster than that of the traditional algorithm when running on CPU.

Key words: H.264/AVC;CUDA;motion estimation;hardware accelerator;heterogeneous computing