Dense Tracking and Mapping in Real-Time

Dense Tracking and Mapping in Real-Time

ID:40370055

大小:6.33 MB

页数:8页

时间:2019-08-01

Dense Tracking and Mapping in Real-Time_第1页
Dense Tracking and Mapping in Real-Time_第2页
Dense Tracking and Mapping in Real-Time_第3页
Dense Tracking and Mapping in Real-Time_第4页
Dense Tracking and Mapping in Real-Time_第5页
资源描述:

《Dense Tracking and Mapping in Real-Time》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、DTAM:DenseTrackingandMappinginReal-TimeRichardA.Newcombe,StevenJ.LovegroveandAndrewJ.DavisonDepartmentofComputing,ImperialCollegeLondon,UK[rnewcomb,sl203,ajd]@doc.ic.ac.uktemsworkingwithahand-heldcamerahaverecentlyap-Abstractpeared(e.g.[9,13]),butthesestillrelyonfeaturetracking

2、.Herewepresentanewalgorithm,DTAM(DenseTrackingDTAMisasystemforreal-timecameratrackingandrecon-andMapping),whichunlikeallpreviousreal-timemonoc-structionwhichreliesnotonfeatureextractionbutdense,ularSLAMsystemsbothcreatesadense3Dsurfacemodeleverypixelmethods.Asasinglehand-heldRG

3、Bcameraandimmediatelyusesitfordensecameratrackingviawholefliesoverastaticscene,weestimatedetailedtextureddepthimageregistration.Asahand-heldcamerabrowsesascenemapsatselectedkeyframestoproduceasurfacepatchworkinteractively,atexture-mappedscenemodelwithmillionswithmillionsofvertic

4、es.Weusethehundredsofimagesofverticesisgenerated.Thismodeliscomposedofdepthavailableinavideostreamtoimprovethequalityofasim-mapsbuiltfrombundlesofframesbydenseandsub-pixelplephotometricdataterm,andminimiseaglobalspatiallyaccuratemulti-viewstereoreconstruction.Thereconstruc-regu

5、larisedenergyfunctionalinanovelnon-convexopti-tionframeworkistargetedatourlivesetting,wherehun-misationframework.Interleaved,wetrackthecamera’sdredsofnarrow-baselinevideoframesaretheinputtoeach6DOFmotionpreciselybyframe-ratewholeimagealign-depthmap.Wegatherphotometricinformatio

6、nsequentiallymentagainsttheentiredensemodel.Ouralgorithmsareinacostvolume,andincrementallysolveforregularisedhighlyparallelisablethroughoutandDTAMachievesreal-depthmapsviaanovelnon-convexoptimisationframeworktimeperformanceusingcurrentcommodityGPUhardware.withelementsincludinga

7、cceleratedexactexhaustivesearchWedemonstratethatadensemodelpermitssuperiortrack-toavoidcoarse-to-finewarpingandthelossofsmalldetails,ingperformanceunderrapidmotioncomparedtoastateofandaninterleavedNewtonsteptoachievefineaccuracy.theartmethodusingfeatures;andalsoshowtheadditionalM

8、eanwhile,inaninterleavedfashion,thecamera’sposeisusefu

当前文档最多预览五页,下载文档查看全文

此文档下载收益归作者所有

当前文档最多预览五页,下载文档查看全文
温馨提示:
1. 部分包含数学公式或PPT动画的文件,查看预览时可能会显示错乱或异常,文件下载后无此问题,请放心下载。
2. 本文档由用户上传,版权归属用户,天天文库负责整理代发布。如果您对本文档版权有争议请及时联系客服。
3. 下载前请仔细阅读文档内容,确认文档内容符合您的需求后进行下载,若出现内容与标题不符可向本站投诉处理。
4. 下载文档时可能由于网络波动等原因无法下载或下载错误,付费完成后未能成功下载的用户请联系客服处理。