/
ferentlightingconditions,andfromdifferentdistances–alladdingtothe ferentlightingconditions,andfromdifferentdistances–alladdingtothe

ferentlightingconditions,andfromdifferentdistances–alladdingtothe - PDF document

conchita-marotz
conchita-marotz . @conchita-marotz
Follow
365 views
Uploaded On 2016-07-04

ferentlightingconditions,andfromdifferentdistances–alladdingtothe - PPT Presentation

For a sense of the ac cu racy and re al time re spon sive ness of the pro to type we in vite the reader to watch the fol low ing video demon stra tion of our live sys temhttpsynrgcslillinois ID: 390761

For a sense of the ac cu racy and re al time re spon sive - ness of the pro to type we in vite the reader to watch the fol low ing video demon stra tion of our live sys tem.http://synrg.csl.illinois.

Share:

Link:

Embed:

Download Presentation from below link

Download Pdf The PPT/PDF document "ferentlightingconditions,andfromdifferen..." is the property of its rightful owner. Permission is granted to download and print the materials on this web site for personal, non-commercial use only, and to display it on your personal computer provided you do not modify the materials and that you retain all copyright notices contained in the materials. By downloading content from our website, you accept the terms of this agreement.


Presentation Transcript

ferentlightingconditions,andfromdifferentdistances–alladdingtothecomplexityoftheproblem.Ifthedatabasecon-tainsmanyimagesofthesameobject(e.g.,Aliceandmanyothershaveannotatedthesamepainting),imagematchingexploitsthisdiversitytogainrobustness.However,ittakeslonger.Withonlyoneorfewimagesinthedatabaseperob-ject,theoperationisfaster,butatthecostofaccuracy.Recentyearshavewitnessedhybridapproachesthatfusesen-sordataandcomputervision[24,28,33].Byutilizingthelocationoftheuserandorientationofthephone,authorsin[38]prunethesearchspaceforimagematching.Thiscer-tainlyoffersimprovement,however,thelackofpreciselo-cationandlargecompassuctuationsmakethemunreliableindoors.Thelatencyisalsointheorderofseveralseconds,requiringuserstotakeapictureandwaitfortheannotation.Real-timeobjectbrowsing–likeamagnifyinglens–reallywarrantssub-secondlatency.Basedonextensiveliteraturesearch,weconcludethatde-spitemeaningfultechnicaladvances,andmanyreleasedappsandmediaarticles,nosingleefforthasfullydeliveredonthecompletevision.Thereareholesintheendtoendprocess-ingpipeline–achievingthestrictlatencyandaccuracytargetswouldwarrantaholisticlookatthesystem.Thesystemalsoneedstorelaxanyassumptionsonlocationtobedeployableuniversallyinthenearfuture.Finally,sheerengineeringisnecessarytocopewithcornercasesinuncontrolledenviron-ments,includinghandvibrationsinsomeusers,someobjectsbeingidentical,weaknetworkconnections,etc.Weaimtocompleteandtestthevision–todeliver,fromatop-downdesign,aviablemobileARframeworkthatenablesauseableandrewardingexperience.Whilemanyofthechal-lengesthatmustbeaddressedarenotfundamental,theyare“hard”andpreviouslyundemonstrated.Itneededmanyde-signiterationstogetthepipelinesufcientlyoptimizedsothataviablesolutioncouldevenbewithingrasp.Initially,weexpectedoureffortstobeprimarilyengineering.However,wealsofoundripeopportunityfornovelheuristicstoimproveaccuracy,responsiveness,andultimately,thehumanuserexpe-rience.Especially,wedevelopedanoptimizationframeworkthatunderpinscomputervisionwithageometricrepresenta-tionofobjectsinthesurroundings.Weareabletolearnthatgeometryfromthenaturalhumanmovementsacrossspaceandtime,bringingthesystemtoadesirablelevelofrobust-ness.Finally,weevaluateourapproachnotonlythroughmicrobenchmarks,butalsolive,throughthenaturalusageofin-dependentandunbiasedvolunteers.Forthersttime,wewereabletoobserverealhumanbehaviorswheninteractingwithaviablemobileARsolution.Indevelopingoursolution,welearnedthekeycontributionofthispaper:MobileARisbestaddressednotthroughsensing,computervision,oranymodalityinisolation.Instead,eachmaybemostadvantageouslycombinedwithinastatefulmodel,underpinnedbythegeometryandtimeofmotion.Withthisspatiotemporalawareness,itspossibletodelivercontenttothehumanuser,atconsistentprecisionandsub-secondlatency. For a sense of the ac cu racy and re al time re spon sive - ness of the pro to type, we in vite the reader to watch the fol low ing video demon stra tion of our live sys tem.http://synrg.csl.illinois.edu/projects/MobileAR2.MEASUREMENTSANDGUIDELINESOverLayaimstoenablerealtimeaugmentedrealityonto-day'ssmartphoneplatforms.Thissectiondevelopsbasicde-signguidelinesthroughmeasurementsandobservations.LocationandOrientationSensorsInadequateWebelievethatsensor-onlyapproachesareinadequatetore-alizeOverLayinanygeneralizedsettings.Eveninoutdoorsettings,whereGPSlocationsareprecisetoaround5m,sen-sorbasedapproachesfallshort.Figure1showsmeasure-mentsperformedwithWikitude[9],apopularappontheappstorethatusesthephone'slocation,compass,andaccelerom-etertodisplayannotations.ThegraphplotsWikitude'serrorinlineofsights(LoS)measuredbycomputingtheperpen-diculardistancefromthetrueobjecttotheLoS.Themedianseparationisaround12m,implyingthatanobjecthastobeatleast12mwidefortheannotationtostillbecorrect.Clearly,thisofferslimitedapplicability. Figure1:DifferenceinestimatedLoSvs.groundtruthwhenmeasuredthroughtheWikitudeapp.Indoorenvironmentsarefarworse.Figure2measurescom-passerrorsduetoferromagneticmaterialintheambience–asauserwalksthroughastraightcorridor,compassangledeviationsaremeasuredagainstgroundtruth.Themediandeviationismorethan15withaheavytailofupto180.Finally,indoorlocationisstillnotuniversallyavailable,andsomethatarestartingtorolloutinafewplacesarelimitedto5maccuracies,atbest.Suchaccuracieseasilyderailaaug-mentedrealityapproach.Thus,tomakeOverLayrobustandimmediatelydeployableinallenvironments,wemustdesen-sitizethesolutiontolocalizationandorientationestimates. Figure2:Distributionofcompassdeviationinvariousin-doorenvironments.ComputerVisionandCloudOfoadEssentialTherealityweintendto(digitally)augmentisvisual.Inlightofthis,itseemsmostintuitivetotakeadvantageofthecam-era,asensorthat“sees”theworldsimilartohumans1.Of 1Weassumecurrentgenerationcamerasinsmartphones.While3Dcameraswouldopenfurtheropportunities,wewishtoensurethewidestgeneralizabilityacrossdevicestoday.

Related Contents


Next Show more