| Video coding is the process of converting an input video sequence into a binary stream through a series of compression algorithms.With the increasing complexity of video coding and the growing demand for ultra-high definition videos,the average duration of encoding a video sequence also increases significantly.In this paper,the research on Versatile Video Coding(VVC)and Audio Video Standard II(AVS2)are undertaken to reduce encoding time,using GPU acceleration and transcoding acceleration,etc.This article contains two parts of research content.The first part is GPU-based sub-pixel interpolation acceleration for VVC.In this section,the complexity of VVC and High Efficiency Video Coding(HEVC)are compared.The design process of interpolation kernel function is given based on the hardware characteristics of GPU and CUDA programming model.The kernel functions are optimized for memory addressing,shared memory access,thread block size,and input image size.The experimental results show that the sub-pixel interpolation accelerated by GPU is 67.12% faster than the original instruction set acceleration,and the total encoding speed is 6.35% faster than the anchor,while the coding performance remains unchanged.The second part is the research of transcoding acceleration for AVS2.Considering the similarity between HEVC and AVS2,and the large demand of AVS2 in broadcasting industry,this paper implements a HEVC to AVS2 transcoder that supports parallel transcoding,where multiple levels of configuration are given to balance speed-up and performance.The experimental results show that compared with single-threaded full-code transcoding,the transcoder proposed in this paper is 11.07 times faster in single-thread mode,and 66.93 times faster in multi-thread mode.Considering the current supply and demand relationship and market price of AVS2 real-time encoder,this paper further cascades the existing HEVC real-time encoder in front of this transcoder,which can be used as AVS2 fast encoder at low cost. |
PDF Full Text Request