FraunhoferIISUseCasePolarizationImageProcessingSystem2017/01/12

• NovelpolarizationimagesensorandcamerasystemdevelopedatFraunhoferIIS• Polarization:physicalpropertyoflight(likewavelengthandintensity),butnot

perceivabletohumans• Usedinindustrialapplicationslikequalitymonitoringinglassorcarbonfibre

production• Medicalapplicationsarecurrentlyinresearch• Currently,PCsoftwareforimageprocessingisavailable;imageanalysisishighly

applicationdependentandcustomerspecific• Softwaredevelopmentinthetraditionalway:Matlabsimulation,conversionto

C/C++,optimizationfortargetarchitecture,debugging,profiling…• ARGOworkflowwilldrasticallyreducetime-to-market(factorof3...5expected)• Inthefuture,partsofthealgorithmwillbeimplementedonembeddedsystems

(insidethecamera)Keywords:Worstcaseexecutiontime,embeddedsystem,safetycritical,modelbaseddesignThisprojecthasreceivedfundingfromtheEuropeanUnion’sHorizon2020researchandinnovationprogrammeundergrantagreementNo688131

ChallengesPolarizationimageprocessingsystemisusedastestcaseintheARGOproject

• Testcaseincrement1containsmostcommonprocessingsteps(commontoalmostallapplications)

• Testcase2willcontaincomputationintensiveenhancementsandapplicationspecificprocessingsteps.Morechallengingw.r.t.WCETconstraints

Solution

• ModelintheLoop:SimulationPC• HardwareintheLoop:ARGOplatform

Phase1:Infrastructurebuildupandtest:

• ImplementScilabmodelandinfrastructuresoftware,hardwaresetup• Recordshortsequencesofrawdata(=testdata)fromthepolarizationcamera

(~2000framespersequence,~50sec)• ProcesstestdataintheScilabPIPSmodel,eachresultinginaAOMPandDOLP

sequence(=referencedata)Phase2:ModelintheLooptest

• Transmitpreviouslyrecordedtestdatasequencesvia EthernetframewisetoasimulationPC(alsorunningthePIPSmodel)andreceiveprocessedframes

• CompareresultsfromsimulationPCtoreferencedata.Theymustbeexactlyidentical.

• NoWCETconstraintsapplicableinthisphasePhase3:Hardwareinthelooptest:(byendoftestcase1)

• TransmittestdatasequencesviaEthernetframewisetotheARGOtargethardware(runningthecompiled andWCET-optimizedPIPSmodel)andreceiveprocessedframes.

• Useincreasingframeratesforstressingthetargetsystem.• Resultsfromthetargetshouldbeidenticaltoreferencedataexceptforround-off

effectscausedbyarithmeticprecisiononthetarget.

DetailsoftheScilabPIPSModelItcoversthemostcommonstepsusedinalmosteverypolarizationimagingapplication.

OutlookforTestcaseIncrement2• Polynomialbasedpixellinearizationforhigherprecision(Note:cameramustbe

seenasameasuringinstrumentratherthanavisualimagesource)

• Temperaturecompensation• Applicationspecificprocessing,e.g.ATN(apperenttempernumber)• Enhanceddenoising(↔unweightenedframeaveraging)• Edge-awarepixelinterpolation(↔simpleneighbouraveraging)

ResultsWhatisthebenefitweexpectfromARGO?

• Drasticallyreducedtime-to-market• Reducednumberofiterationsinsoftwaredevelopmentcausedbyerrornous

C-Conversionofsimulationcode• Higherperformanceofresultingimplemenationduetoautomatic

parallelizationandWCETawareness→Moreapplicationareas• Mucheasiermigrationtoothertargetplatforms,i.e.embeddeddevices

Example:ModelintheLoopusingRaspberryPiassimulationplatfom

Moreinfoshttps://www.iis.fraunhofer.de/en/ff/bsy/tech/kameratechnik/polarisationskamera.htmlhttp://scilab.io/use-cases/real-time-embedded-systems-require-higher-performance/