資源描述:
《頭戴式視線跟蹤系統(tǒng)關(guān)鍵技術(shù)研究》由會(huì)員上傳分享,免費(fèi)在線閱讀,更多相關(guān)內(nèi)容在學(xué)術(shù)論文-天天文庫(kù)。
1、摘要摘要注視信息反應(yīng)了人的感興趣點(diǎn),在人機(jī)交互中可以通過注視輸入信息;同時(shí),眼睛也是接收信息的主要通道,因此利用視線作為人機(jī)交互接口,會(huì)使人機(jī)交互更加自然和方便。視線跟蹤技術(shù)是未來智能人機(jī)交互接口的關(guān)鍵技術(shù)之一,在游戲、虛擬現(xiàn)實(shí)和智能計(jì)算,軍事領(lǐng)域等都有著廣泛的發(fā)展前景。本文研究頭戴式視線跟蹤系統(tǒng)中的關(guān)鍵技術(shù),主要工作和研究成果如下:1、分析了頭戴式視線跟蹤系統(tǒng)標(biāo)定過程的干擾樣本問題,采用隨機(jī)采樣一致性算法剔除樣本集中的噪聲點(diǎn),保證了標(biāo)定結(jié)果的準(zhǔn)確性。標(biāo)定過程是視線跟蹤一個(gè)關(guān)鍵步驟,用于建立眼睛圖像與場(chǎng)景圖像的映射
2、關(guān)系。由于人的眨眼動(dòng)作、跟蹤目標(biāo)突然改變位置時(shí)眼睛反應(yīng)的滯后性、視線不集中等因素,導(dǎo)致標(biāo)定過程采集的樣本集中存在干擾樣本,這些干擾樣本會(huì)嚴(yán)重影響所訓(xùn)練的映射矩陣的精度。本文采用隨機(jī)采樣一致性算法估計(jì)映射矩陣,實(shí)驗(yàn)結(jié)果表明,該方法可以有效地消除樣本集中干擾樣本的影響。2、實(shí)現(xiàn)了標(biāo)記點(diǎn)檢測(cè)和平面幾何映射的注視點(diǎn)計(jì)算方法。在估計(jì)出場(chǎng)景圖像中注視點(diǎn)后,通過實(shí)時(shí)檢測(cè)場(chǎng)景圖像中標(biāo)記點(diǎn)的位簧,構(gòu)建場(chǎng)景圖像和計(jì)算機(jī)屏幕的平面映射關(guān)系,將場(chǎng)景圖像中的注視點(diǎn)映射到屏幕上。實(shí)驗(yàn)結(jié)果表明該方法簡(jiǎn)單易行,可以較好的估計(jì)出用戶在真實(shí)場(chǎng)景中的注
3、視點(diǎn)坐標(biāo)。關(guān)鍵詞:視線跟蹤標(biāo)定算法隨機(jī)采樣一致性算法注視點(diǎn)估計(jì)瞳孔定位AbstractTheperson’Sgazeinformationreflectswhoseinterestingareaandcanbeusedasinterfkeforhumancomputerinterction.Simultaneously,eyesarethemainchanneltoreceivetheoutsideinformation,SOthegazeusedastheinterfacemakeshumancomputerin
4、terctioneasyandnatural.Eyetrackingisoneofthemostimportanttechniquesforfutureartificialintelligence.Thekeytechniquesofhead—mountedeyetrackingsystemwerestudiedinthisthesis.a(chǎn)ndthemainworksandresearchresultsaresummarizedasfollows:Jj確e西曰H翎ceofoutliersmixedindatasam
5、plescollectedduringcalibrationisstudiedsystematically,andRandomSampleConsensusisemployedtoprunetheoutliers.Calibrationisakeyprocedureforeyetrackingsystem·Thepurposeofcalibrationistoconstrutedtheeyegazemappingfunctionfromeyeimagecoordianatetosceneimagecoordiana
6、te.However,inpracticethedatasamplecollectedduringcalibrationmaybecorruptedbysomeoutliersbecauseofblinkactionorsomeotherreasons.Theseoutlierswillbadlyinfluencetheaccuracyofmappingfunction·AndRalldomSampleConsensusisemployedtoestimatethemappingfunctionwithdatas鋤
7、pleswhichmayincludesomeoutliers.Experimentalresultsshowthatthemethodcanimprovetheaccuracyofmappingfunctiondramatically.2.AgeometrybasedmethodiSproposedtoestimatethegazeofpointinscreenforhead.mountedeyetrackingsystem.Somemarkersaresetaroundthescreen,andtlleposi
8、tonsofthesemarkersinsceneimagearedetectedinrealtime·Therelationshipbetweensceneimageandcomputerscreenisconstructedwiththesepointcorrespondencesintwoviews.Thepointofgazeinthesceneim