基于Haar-like特征的實時道路車輛識別方法研究.pdf

1、青島大學碩士學位論文基于Haarlike特征的實時道路車輛識別方法研究姓名：張亮修申請學位級別：碩士專業(yè)：車輛工程指導(dǎo)教師：王玉林20090609AbstractRealtimeonroadvehicledetectionisakeytechnologyinmanytransportationapplications，suchasdriverassistance，autonomousdrivingandactivesafetyDueto

2、varietiesinvehicleappearanceandcomplexitiesofoutdoorenvironment，itisimportanttodistinguishvehiclesfromotherobjectsasrobustandreliable嬲possiblenesupervisingapproachofstatisticalpatternrecognitionisutilizedinthisthesistotr

3、ainaclassifierbasedonHaarlikefeaturewhichimplementsrealtimeonroadvehicledetectionBasedonpreviousresearch，thestructuresofHaarlikerectangularfeaturesandintegralimagethatcanbeusedtocomputethevalueoffeaturesrapidlyareanalyze

4、dElevensortsofexpandedrectangularfeaturesthatLienhartRputforwardareadoptedtepresentvehiclesInaddition，thispaperdevelopasinglefeaturetepresenttheshadowatthebottomofvehicles，andconstructUtypefeatureusingtwoverticaledgesand

5、shadowofvehicleHaarlikefeaturesareusedtoconstructweakclassifiers，andGentleAdaBoostalgorithmischosentotrainthestrongclassifiersThen，thereal—timeon—roadvehicleclassifierbasedontreestructureisconstructedbycombiningthestrong

6、classifiers，whichalsoUSeSclusteringand—splittingsteprecursivelyandtherootnodeoftreeclassifierisastrongclassifierthatemploysUtypefeatureMultipledetectionwindowapproachisintroducedtodetectvehiclesinimages，avoidingscalingim

7、agedirectlyintraditionalwayTreeclassifierrapidlyrejectsasmanynegativesaspossibleattheearliestlayers，onlytruepositiveareasreachtheleavesterminalstrongclassifiersThedatasetswhichconsistoftrainingimagesandtestingimagesareco

8、llected，andvideodatasetiScollectedforrealtimevehicledetectionTherealtimeonroadvehicleclassifierbasedontreestructureistrainedusingtrainingdatasetandevaluatedintestingdatasetandvideodatasetExperimentalresultsshowthatthetre