基于聚类的背景建模与运动目标检测方法.pdfVIP

, () ComputerEngineeringandApplications计算机工程与应用 2008448 193 基于聚类的背景建模与运动目标检测方法 李庆忠,何东晓 , LIQing-zhongHEDong-xiao 中国海洋大学 工程学院,山东 青岛 266100 , , , , CollegeofEngineeringOceanUniversityofChinaQingdaoShandong266100China : E-maillqzhlz@ , LIQing-zhongHEDong-xiao.Newmethodforbackgroundmodelingandmovingobjectdetectionbasedonclustering. , , (): ComputerEngineeringandApplications2008448 193-195. : Abstract Inordertoreducethecomputationaltimeandmemoryrequirementoftheexistingdynamicbackgroundmodeling , methodsthispaperpresentsanewbackgroundmodelingandmovingobjectdetectionmethodbasedonclustering.Foradynamic , , backgroundthehistogramofeachpixelvalueovertimeisusuallyintheformofmultimodal.Thereforeregardingeachpeakasa , clusterclusteringtechniqueisemployedtoconstructandupdatethemodelofadynamicbackground.Thenbyusingtheestab- , lishedbackgroundmodelthemovingobjectsaresegmentedfromthebackgroundquicklyandaccurately.Experimentalresultsshow , thattheproposedbackgroundmodelingmethodcaneffectivelycaptureandadapttothechangesinbackground.Inadditionthis , methodoutperformsthecurrentbackgroundmodelingmethodsintermsofcomputationaltimeandmemoryrequirementthusbe- comingeasytoimplementforDSPorFPGAbasedhardware. : ; ; ; Keywordsbackgroundmodelingclusteringmovingobjectdetectionvideosurveillance 摘 要:为克服目前动态背景建模方法中计算量和存储量大的问题,提出了一种基于聚类的动态背景建模与运动目标分割方 法。由于动态背景下每个像素的取值在时间轴上呈多峰分布形式,因此将每个峰看成一个子类,用聚类技术快速实现了动态背 景的建模与更新,然后利用建立的背景模型快速、准确地实现运动目标的分割。实验结果表明:提出