| UHD(Ultra high definition)video is a new round of major technological innovation following the digitization and high-definition of video,which will drive profound changes in all links of the industry chain.However,the huge amount of ultra-high-definition video data puts tremendous pressure on its transmission,storage and processing.In the case of limited bandwidth,the transmission frame rate of video is often 30 Hz or 60 Hz.However,due to the differences in playback devices,mainstream display devices generally have a refresh rate of 60 Hz,and a few high-end products have screens with a higher refresh rate.Due to the difference in the refresh rate between the source and the screen,there will be phenomena such as smear,pause,blur,etc.,so that the video we watch cannot offer high quality playback.If the video transmission frame rate or resolution is increased according to the requirements of the device,it will result in a significant increase in bandwidth.This leads to an imbalance between image quality and transmission bandwidth in terminal systems with different performance.This article first discusses the solution of using end-cloud fusion for video processing,moving the motion estimation originally located in the terminal to the cloud.In this way,the redundant information of the video image can be removed in the cloud,and the large-scale frame loss can be realized,thereby reducing the transmission bandwidth required for the video,and can make full use of the advantages of cloud computing performance to calculate more accurate motion estimation information than the original terminal's calculation results.Then,perform smoother frame insertion on the terminal to achieve better playback effects.Next,this article designs the use of hierarchical motion estimation technology in the cloud to perform the bidirectional motion estimation algorithm at the low-resolution layer.The algorithm is based on the completion of forward scanning and reverse scanning under the low resolution layer,so as to provide more accurate motion vectors for the 8K resolution layer.The algorithm mainly includes the selection of candidate vectors,the generation of penalty values,the increase of fine search modes,and the selection of scanning methods.Subsequently,this article designs the overall scheme of the logic implementation part of the algorithm and the detailed scheme of each sub-module,as well as a series of optimizations based on the overall motion estimation system for this module under the requirements of time-sharing multiplexing,and the reuse of the cache can be realized by reasonably splitting the cache and reasonably distributing the data in the cache.Finally,the circuit synthesis and functional verification were completed through the Cadence software platform,and the code coverage rate was used to ensure the completeness and effectiveness of the verification.Then the DC synthesis results and power consumption simulation results were given and analyzed,and from the perspective of module and cache reuse to illustrate the module's achievability.The circuit implementation part of this article uses SMIC's 28 nm process,the bus frequency is 200 MHz,and the working clock is 400 MHz.Through the use of layered motion estimation technology to complete the two-way motion estimation under the 1920×1080 low-resolution layer,the forward and backward scanning of each frame of image takes about 6 ms in total.The total area of the logic part and the cache part is 287625 ?m~2,and the peak power consumption is 33.05 m W. |
PDF Full Text Request