Fast Motion Estimation Of H.264 And The Implement On CUDA

H.264 is a new generated standard of video compression which formulated by two important international standard organizations, ISO/IEC and ITU-T. In H.264, the coding efficiency is enhanced by a wide margin. However, the computational complexity is increased efficiently. It's a big challenge for real-time video encoding. The standard can be improved from two aspects. On the one hand, the encoding speed could be increased by improving the coding algorithm, on the other hand, we use co-processing equipment to work with CPU together.Motion estimation is one of the core techniques of H.264 video coding. It takes a large proportion of coding time. This paper has studied several classic motion estimation algorithm with detailed description and analysis. Then, an improved UMHEX fast motion estimation algorithm is proposed. In the algorithm, a series of techniques are adopted, such as early termination, improved search template and mixed search template. The result shows that compared with Full Search algorithm and UMHEX algorithm, the improved algorithm can enhance the processing speed while PSNR is slightly degraded and the increasing bit rate is negligible.Taking the advantage of powerful floating-point computation and parallel ability of GPU, we can use GPU to do video encoding with CPU together. In video decoding, the deblocking filter module is shifted to GPU to accelerate. A parallel encoding architecture CPU+GPU is proposed. The deblocking filter time is reduced greatly by full use of the GPU's parallel computing power. The result shows that compared with traditional algorithms, the parallel architecture can effectively accelerate the encoding. Both the improvement on the algorithm and the using of co-processor can effectively improve the coding speed, so that H.264 video encoding standard can be well applied to the actual occasion.