计算机与现代化

• 图像处理 • 上一篇    下一篇

一种基于SURF与地理格网模型的增强现实方法

  

  1. (交通运输部天津水运工程科学研究所,天津300456)
  • 收稿日期:2018-02-27 出版日期:2018-07-05 发布日期:2018-07-05
  • 作者简介:毕金强(1985-),男,山东威海人,交通运输部天津水运工程科学研究所工程师,硕士,研究方向:GIS空间分析,图像处理。
  • 基金资助:
    中央级公益性科研院所基本科研业务费专项资金资助项目(TKS160213)

AnAugmentedRealityMethodBasedonSURFandGeographicGridModel

  1. (TianjinResearchInstituteforWaterTransportEngineering,M.O.T.,Tianjin300456,China)
  • Received:2018-02-27 Online:2018-07-05 Published:2018-07-05

摘要: 图像识别与匹配是增强现实领域研究与应用的基础和关键,针对户外场景的广域性和随机性,以及目标纹理结构相似性等问题,提出一种基于SURF与地理格网模型的增强现实方法。该方法根据目标场景与地理位置的相关性,检测图像特征点并生成Location-SURF图像特征描述,基于地理格网模型构建空间四叉树索引,建成静态特征样本库。将视频帧、位置和角度信息生成特征图像,上传至服务端解析运算并与样本库训练匹配。选取宁波环球航运广场约0.376km2的区域,采集270余幅图像数据构建样本库并开展试验,通过现场图像的实时采集和计算,能够实现特征点的在线匹配,在此基础上通过调整点位距离比例的阈值,能够提升匹配的准确程度。基于该算法开发移动增强现实系统,运用四层技术架构实现了终端采集显示和服务端分析计算的并行化,达到真实场景与虚拟信息的融合显示效果。系统应用结果表明:该算法可以解决复杂环境下场景图像识别匹配率不高的问题,可快速地完成特征点的检测和提取,能够有效地进行样本训练和匹配,对户外移动增强现实进行了有益尝试并提供一种有效的途径。

关键词: 移动增强现实, 图像识别与匹配, 特征点检测, 地理网格

Abstract: Imagerecognitionandmatchingisthefoundationandkeyofresearchandapplicationonthefieldofaugmentedreality.Accordingtothewideareaandrandomnessofoutdoorscenes,similarityoftargettexturestructure,anaugmentedrealitymethodbasedonSURFandgeographicgridmodelisproposed.ThemethodaccordingtothecorrelationbetweenthetargetsceneandthegeographicallocationdetectstheimagefeaturepointstogeneratetheLocation-SURFimagefeaturedescription,constructsthespatialfourforktreeindexbasedonthegeographicalgrid,andbuildsthestaticfeaturesampledatabase.Featureimagesaregeneratedbyvideoframe,locationandangleinformation,anduploadedtotheservertoparseoperations,andtrainedtomatchwiththesampledatabase.Testsresearchwhichgathers270piecesofimagedatatoconstructasampledatabaseisdoneintheNingboglobalnavigationsquareabout0.376km2.Onlinefeaturepointscanbematchedonlinethroughthereal-timecollectionandcalculationofspotimage,andtheaccuracyofmatchingupcanbeimprovedbyadjustingtheproportionofdistancethreshold.Amobileaugmentedrealitysystemisdevelopedbasedonthismethod,whichusesfourlayersoftechnologyarchitecturetorealizetheparallelizationoftheterminalcollection,displayandserveranalysisandcalculation,andachievesthefusioneffectoftherealsceneandthevirtualinformation.Theapplicationresultsshowthat:Thismethodcansolvetheproblemoflowmatchingofimagerecognitionincomplexenvironment,canquicklycompletethedetectionandextractionoffeaturepoints,caneffectivelycompletesampletrainingandmatching.Aswellasthismethodmakesausefulattempttoprovideaneffectivewayintheoutdoormobileaugmentedreality.

Key words: mobileaugmentedreality, imagerecognitionandmatching, featurepointdetection, geographicgrid

中图分类号: