基于双目视觉的稠密立体匹配算法研究

《基于双目视觉的稠密立体匹配算法研究》由会员上传分享，免费在线阅读，更多相关内容在学术论文-天天文库。

1、分类号UDC密级基学号1307010094于双目视觉的稠密立体硕士学位论文匹配算法研究基于双目视觉的稠密立体匹配算法研究柯柯俊山俊山学科门类：理学学科名称：数学西指导教师：戴芳教授安理工申请日期：2016年3月大学万方数据万方数据万方数据西安理工大学硕士学位论文算法平均排名为112.7。最后，针对被动测量技术中，三角测量方法不能充分利用摄像机内参数的缺陷对其进行了改进。我们应用（2）中的S-ELAS视差计算方法，使用KITTI视觉基准数据库中的交通场景图像对改进的三角测量方法进行了测试。测试结果

2、表明，改进的三角测量方法的平均坐标准确率是89.79%，平均距离准确率是96.60%，可见修正的距离计算方法非常精确。关键词：双目视觉；立体匹配；SIFT；SLIC；ELASII万方数据AbstractTitle：STUDYONDENSESTEREOCORRENPONDENCEALGORITHMFORBINOCULARVISIONMajor：MathematicsName：JunshanKeSignature:Supervisor：Prof.FangDaiSignature:AbstractBa

3、sedontheunderstandingofbinocularstereovisionsystem,wehavedonesomeresearchesonfeature-basedandwindow-basedstereocorrenpondencealgorithms,andmainlydiscussedhowtoimprovethedisparityresultofbinocularstereomatchingalgorithmandhowtodevelopdistancecalculati

4、onmethod.Thestereomatchingalgorithmoftenmakeitdifficulttoobtainaccuratedisparitysearchrange,expensivecostofcomputationandlowmatchingrate,thisthesispresentstwoimprovedalgorithms:(1)TakingtheadvantagesofSIFTfeatureextractionalgorithm,EMDalgorithmandwin

5、dow-basedmatchingtechniquesintoconsideration,aimproveddensestereocorrespondencealgorithmisproposed.First,weproposealocalfeaturedetectionalgorithmcombinedSIFTwithEMD.BecauseSIFTisaveryrobustlocalfeatureextractiontechniquewhichisgoodforcapturingthemost

6、importantfeaturesofimage,andtheEMDmethodisself-adaptiveandhighlyefficientinanalyzingnonlinearandnon-stationarysignalandthismethodisabletodecomposethesignalintothesumofsimplesignalsindifferentfrequency.Afterthefeatureextractionofimage,thematchingpoint

7、pairswithhigheraccuracyareobtainedbyfeaturematchingmethodandepipolargeometricconstraint,andthenthesematchingpointsareusedtoestimatetheinitialdisparitysearchrange.Second,onthebasisofnewblockmatchingstrategyandmodifiedmatchingcostfunction,thedensedispa

8、ritymapisobtainedbytheimproveddisparityreliabilityfunctionorweightenergyfunction.TheresultsshowthatthefeaturedetectionalgorithmcombinedSIFTwithEMDismoreaccuratethanthetraditionalSIFTapproach,anditcouldclosertheactualrequirementdisparitysearchrange.An