双目立体匹配的算法分析及其多核并行化-algorithm analysis and multicore parallelization of binocular stereo matching.docxVIP
- 18
- 0
- 约3.76万字
- 约 57页
- 2018-07-31 发布于上海
- 举报
双目立体匹配的算法分析及其多核并行化-algorithm analysis and multicore parallelization of binocular stereo matching
ABSTRACTThestereomatchingalgorithmforbinocularstereovisionisacriticalandchallengeissueinthefieldofcomputervision,whichmayhaslowermatchingqualityorhighertimecomplexity.Toimprovethat,aboundaryextensionispre-processedtoimprovematchingaccuracyoftheimageboundaryarea,offereda5%matchingaccuracyincrease;andthenahierarchicalmulti-resolutionblockmatchingalgorithmforhigh-resolutionimagepairisproposedinthepaper;basedonthehierarchicalmulti-resolutionblockmatching,anovelfastmethodisproposed,whosecomplexityisonlyone-ninthoforiginalalgorithm;andamethodofvariablewindowisadoptedtoobtainmoresatisfactoryresultontheuntexturedregions,offeredafurther2%matchingaccuracyincrease.Stereomatchingalgorithmsarecomputationallyintensivealgorithm,withgreatcomputationalcostandavailableparallelism.Withmulti-coretechnologydeveloping,moreandmoreresearcherspayattentiontousingmulti-coreprocessorsandmany-coreprocessor.Inthispaper,basedontheresearchofNVIDIAGPGPU,theGPGPU-basedparallelstereomatchingalgorithmisproposed,whichcanbe190timesfasterthanthetraditionalCPUimplementation.Keywords:StereoMatching;Support-WeightMatching;HierarchicalBlockMatching;VariableWindow,GPGPU目录摘要IABSTRACTII第一章绪论11.1立体匹配的研究背景与意义11.2国内外立体匹配研究现状21.2.1代表性工作21.2.2立体视觉研究存在的问题和发展趋势31.3本文研究内容及章节安排3第二章双目立体匹配理论概述52.1引言52.2立体匹配的相关理论62.2.1立体匹配前期工作62.2.2立体匹配的基本约束62.2.3立体匹配计算步骤及常见算法分析72.2.4立体匹配的评价112.3局域立体匹配算法132.3.1局域匹配窗口选择132.3.2经典的自适应支持权重立体匹配算法162.4本章小结19第三章改进的自适应加权立体匹配算法203.1基于自适应支持权重匹配的图像边界扩展203.2金字塔模型快速立体匹配算法223.2.1金字塔模型233.2.2基于金字塔模型的自适应加权立体匹配233.2.3实验结果分析253.3快速自适应加权匹配算法273.4改进的区域增长立体匹配算法283.4.1一种基于区域增长的立体匹配算法293.4.3实验结果分析313.5本章小结32第四章加权立体匹配算法并行化实现344.1多核计算平台概述344.1.1并行计算概述344.2立体匹配算法多核计算平台实现414.2.1立体匹配算法并行化分析414.2.2立体匹配算法多核计算平台实现步骤424.2.3立体匹配算法多核计算平台实现结果分析444.4本章小结46第五章总结与展望475.1总结475.2展望47致谢49参考文献50攻读硕士研究生期间发表论文54第一章绪论1.1立体匹配的研究背景与意义视觉是人类观察和认知世界的重要功能手段,是一个古老而重要的研究课题。人类从外界获得的信息约有75%来自视觉系统,用机器模拟人类的视觉功能是人们多年的梦想[1]。现实的世界是三维的空间,然而普通的成像系统只能获取物体的二维信息,在成像时丢失了三维信息。人的一只眼睛所获得的景物图像也如普通的摄像机一般,只能获得二维图像,但是人类通过视觉系统却能很容易地从二维图像中感知三维世界,这都归功于人类视觉系统的高度智能性
您可能关注的文档
- 深层孔隙型热储地热尾水回灌堵塞机理及示踪技术研究——以咸阳回灌二号井为例-research on plugging mechanism and tracer technique of deep porous geothermal tailrace recharge - a case study of xianyang recharge well no.2.docx
- 深度经颅磁刺激电磁场分布的研究分析-research and analysis on electromagnetic field distribution of deep transcranial magnetic stimulation.docx
- 深度神经网络音素识别系统及其隐藏层分析-phoneme recognition system based on deep neural network and its hidden layer analysis.docx
- 深度水解蛋白配方乳对早产儿早期胃肠道功能影响及机制探讨-effect and mechanism of deep hydrolyzed protein formula on gastrointestinal function in premature infants.docx
- 深度视频编码快速模式选择算法分析-analysis of fast mode selection algorithm for depth video coding.docx
- 深度手烧伤患者的预后分析-prognostic analysis of patients with deep hand burns.docx
- 深度学习在图像识别中的分析及应用-analysis and application of depth learning in image recognition.docx
- 深吃水多立柱式平台的气隙响应分析-analysis of air gap response of multi-column platform with deep draft.docx
- 深覆盖层土石坝非线性动态特性研究-study on nonlinear dynamic characteristics of earth-rock dam with deep overburden.docx
- 深度视频信号预处理及其在编码中的应用分析-analysis of depth video signal preprocessing and its application in coding.docx
原创力文档

文档评论(0)