OpenCV在TI达芬奇以及OMAP平台下的移植与功能评估Intoday’sadvancingmarket,thegrowingperformanceanddecreasingpriceofembeddedprocessorsareopeningmanydoorsfordeveloperstodesignhighlysophisticatedsolutionsfordifferentendapplications.Thecomplexitiesofthesesystemscancreatebottlenecksfordevelopersintheformoflongerdevelopmenttimes,morecomplicateddevelopmentenvironmentsandissueswithapplicationstabilityandquality.DeveloperscanaddresstheseproblemsusingsophisticatedsoftwarepackagessuchasOpenCV,butmigratingthissoftwaretoembeddedplatformsposesitsownsetofchallenges.Thispaperwillreviewhowtomitigatesomeoftheseissues,includingC++implementation,memoryconstraints,floating-pointsupportandopportunitiestomaximizeperformanceusingvendor-optimizedlibrariesandintegratedacceleratorsorco-processors.Finally,wewillintroduceaneweffortbyTexasInstruments(TI)tooptimizevisionsystemsbyrunningOpenCVontheC6000™digitalsignalprocessor(DSP)architecture.BenchmarkswillshowtheadvantageofusingtheDSPbycomparingtheperformanceofaDSP+ARM®system-on-chip(SoC)processoragainstanARM-onlydevice.
1