Real Time Illumination for App Based AR P. B.

Abstract

Abstract AchievingrealisticlightinginmobileAugmentedReality(AR)remainsafundamental challenge,especiallyoncommoditysmartphonesthatlackspecializedhardwareorad- vancedrenderingcapabilities.Amongthemostcriticalcomponentsofvisualrealismis shadowbehavior—particularlytheabilitytomatchreal-worldlightingdirectionand projectaccurateshadowsbeneathvirtualobjects.Whenshadowsaremisalignedor poorlyrendered,theillusionofrealismbreaksdown,reducingtheimmersivequalityof the ARexperience. This researchpresentsalightweight,hardware-freemethodforestimatingthedirec- tion ofadominantlightsourceandrenderingdynamicshadowsinapp-basedARenviron- ments.Unliketraditionalmethodsthatrelyondepthsensors,LiDAR,HDRimaging,or pre-trained neuralnetworks,thissystemusesacamera-onlyapproachbasedonbrightest pixel detectionandARCorehittesting.Theenvironmentisinitiallyscannedfor3–4 seconds todetectthebrightestpixelinthegrayscalevideostream,whichisassumedto representthedirectionoftheprimaryindoorlightsource.Asecondhittestisperformed at thatpixeltoretrievea3Dcoordinate,fromwhichalightdirectionvectoriscomputed relativetotheplacedobject’sposition.Thisvectoristhenusedtoprojectshadowsusing a customvector-basedrenderingpipeline. The implementationwascarriedoutusingARCoreandSceneform,targetingmid- range AndroiddevicessuchastheSamsungGalaxyM31.3Dmodelswerepreparedin ‘.sfb‘ format,andshadowprojectionwashandlednativelywithoutrelyingonexternal rendering engineslikeThree.js.Shadowsappearashard-edgedsilhouettes,andareup- dated inrealtimeastheusermovesthecamera,maintainingconsistentalignmentwith the perceivedlightdirection.Despitetheabsenceofsoftshadowingorpenumbraeffects, the systemwasabletodeliveracompellinglevelofvisualcoherenceundercontrolled single-lightindoorenvironments. Evaluationwasconductedthroughauserstudyinvolving16participantswhoob- servedbothaphysicalpinksphereanditsvirtualcounterpartrenderedwithshadowsin the ARapplication.Participantsratedshadowalignment,realism,responsiveness,and perceivedqualitythroughastructuredquestionnaire.Resultsshowhighagreementin responsiveness(averagescore:4.50/5),shadowalignment(4.25/5),andperceivedrealism (4.08/5), confirmingtheeffectivenessofthebrightest-pixelandvector-basedshadowing technique.Open-endedfeedbackrevealedappreciationforthesystem’sspeedandvi- sual grounding,whilealsosuggestingenhancementssuchasaprogressindicatorduring scanning andsoftershadowedges. This researchcontributesadevice-compatible,efficient,andinteractivelightingesti- mation pipelinethatavoidsheavycomputationorrestrictedAPIs.Itbridgesthegap betweenperformanceandrealism,offeringapracticalbaselineformobileARapplica- tions inareaslikedesignpreviews,educationaltools,andprototyping.Thethesisalso identifieslimitations—suchaslackofsupportformultiplelightsources,fixedlighting after placement,andabsenceofambientlighteffects—andoutlinesdetaileddirectionsfor future work,includingsoftshadowsimulation,dynamiclightupdates,andcross-platform deployment.