如果您无法下载资料，请参考说明：
1、部分资料下载需要金币，请确保您的账户上有足够的金币
2、已购买过的文档，再次下载不重复扣费
3、资料包下载后请先用软件解压，在使用对应软件打开
基于CUDA的MPEG2编码技术的研究的中期报告Title：ResearchonCUDA-basedMPEG2EncodingTechnologyAbstract:Inthisresearch,wearestudyingtheMPEG2encodingtechnologywhichisbasedonNVIDIA’sCUDAplatform.Wehavecompletedtheinitialresearchwork,andinthismid-termreport,wewillintroducethemainprogressofourresearch.Firstly,weresearchedtheencodingtechnologyandtheCUDAprogrammingmodel.TheMPEG2encodingtechnologyadoptsahybridencodingalgorithm,whichisrelativelycomplex.Weanalyzedthealgorithmanddivideditintoseveralmodules,suchasmotionestimation,discretecosinetransform(DCT)andquantization.ThenweresearchedhowtoacceleratethesemodulesusingCUDAprogramming,whichisaparallelcomputingplatformdevelopedbyNVIDIA.Secondly,weimplementedthemotionestimationmoduleusingCUDA,whichisoneofthemostcomputation-intensivemodulesofMPEG2encoding.Weoptimizedthesearchalgorithmandrealizedthedataparallelismofthecomputingprocess.WecomparedtheperformancebetweenCPUandGPU,andtheresultshowedthatourCUDAimplementationcouldachieve38timesspeedupcomparedwiththeCPUimplementation.Thirdly,westartedtoimplementtheDCTandquantizationmodulesusingCUDA.IntheDCTmodule,weusedthedistributedcomputingofCUDAtospeeduptheprocess.Inthequantizationmodule,weutilizedthesharedmemoryofCUDAanddesignedanewquantizationmatrixtoachievebettercompressionperformance.Finally,weevaluatedtheperformanceofourCUDAimplementationofMPEG2encodingtechnologywithvarioustestdatasets.TheexperimentresultdemonstratedthattheCUDAimplementationcouldachievehighcompressionqualityandsignificantlyincreasethespeedoftheencodingprocess.Inconclusion,ourresearchdemonstratesthatNVIDIA’sCUDAplatformprovidesafeasiblemethodforacceleratingcomplexalgorithmssuchasMPEG2encoding.Inthenextfewmonths,weplantocompletetheimplementationoftheremainingmodulesofMPEG2encodingandevaluatetheperformanceofthecompleteCUDA-basedMPEG2encodingsystem.Keywords:MPEG2encoding,CUDA,motionestimation,DCT,quantization,performanceevaluation.
立即下载
