| In recent years,with the explosive growth of various smart devices(such as smart phones,tablets,VR devices)and the rapid development of wireless network communication technology,mobile video applications,including video streaming providers(such as YouTube,Hulu,Netflix)and socially oriented video applications(like TikTok,Google Hangouts,WeChat)are becoming more popular than the traditional audio and video industry.According to Cisco latest report,as of July 2021,55.56%of web traffic came from mobile devices,and one of the main activities of mobile users is online video,which is expected to grow to 86%by 2023,representing 1 million minutes of video per second content through the network.However,the main challenges in using video codec technology on mobile devices are:1)With the explosive growth of multimedia applications and users’ increasingly high requirements for video quality,how to effectively allocate computing resources(CPU and GPU)has become an important issue;2)High CPU utilization on mobile devices can easily cause the system to become unresponsive;3)Due to the limited battery power of mobile devices,and as users have higher and higher requirements for video resolution,video frame rate and real-time performance,how to minimize the energy consumption of mobile devices while meeting user needs is another major challenge.In this paper,we carry out research work based on the challenges of using video codec on mobile devices,and propose a GPU-based low-power and high-efficiency video codec.The contributions are as follows:1)For the optimization of data transmission between main memory and video memory in the system,this paper proposes frame data management mechanism based on marking syntax.Since the architectural characteristics of mobile GPUs are different from desktop GPUs,the data transmission time between main memory and video memory is unacceptable relative to the real-time performance of video,which has become one of the main bottlenecks in the parallelization of video encoding and decoding on mobile devices.The frame data management mechanism based on syntax mark proposed in this paper adopts data frame buffer technology to eliminate the time consumption of data transmission between main memory and video memory.At the same time,a strategy for frame label classification and an update algorithm are proposed to select appropriate data frames.The experimental results show that,compared with traditional video codecs,the frame data management mechanism proposed in this paper has obvious performance improvement for videos of different resolutions.2)For the inefficient prediction of inter-macroblocks in traditional codecs,a prediction macroblock selection algorithm based on quantization parameter is proposed in this paper.The algorithm proposes a model for prejudging blocks of all-zero discrete cosine transform coefficients.At the same time,combined with the relationship between the integer transform and quantization parameter of video encoding and decoding,the condition for satisfying all zero coefficients of macroblock DCT is proposed,and an improved algorithm for macroblock selection is proposed on this basis.Through experiments to test its performance,the results show that the macroblock selection algorithm proposed in this paper can significantly reduce the complexity of search and improve the efficiency of predicted macroblock selection without significantly reducing the video quality(PSNR and video bit rate).3)For the most time-consuming inter-frame prediction module in traditional codecs,this paper proposes a GPU-based parallel motion estimation algorithm.In order to obtain better parallel performance for encoding and decoding frames,the algorithm proposed in this paper is parallelized in units of macroblocks of each frame instead of pixel-level parallelization.At the same time,the efficiency of matching to the best motion vector can be effectively improved by determining the initial motion vector,coarse-grained search and fine-grained search.This paper conducts experimental evaluations on different GPUs on different mobile devices.The experimental results show that compared with the traditional H.264 codec and some existing improved codecs,the parallel algorithm proposed in this paper not only improves the codec performance significantly,but also can effectively reduce power consumption and CPU utilization. |
