基于运动轨迹分量的行人徘徊行为检测研究

针对传统的目标徘徊检测方法在实时性和准确性等方面的不足,本文提出了一种基于目标轨迹分量曲线的行人徘徊检测算法。首先采用帧差法的背景模板建模方法来建立初始背景。然后用改进的结合背景差分的三帧差分法检测前景目标,通过Mean-shift算法对前景目标进行跟踪。最后将得到的运动轨迹做正交分解,根据根据轨迹的X,Y轴分量曲线来对徘徊行为进行识别。实验表明,该方法能够对几种典型的徘徊行为进行实时、精确判断,同时可以检测出其他复杂的徘徊行为,有较好的实时性和准确率。

WanderingSBehavior Detection Based on PedestrianSTrajectory Component

TraditionalStargetShoveringSdetectionSmethodsShaveSmanySlimitsSandSshortagesSin real-time and accuracy,In this paper , aShoveringSdetection algorithmSbased on the curve ofSpedestrianStarget trajectoryScomponent was proposed.Firstly, the initial background was established with background subtractionStemplateSmodeling method.Secondly,the moving targets were detected by theSimprovedScombined with backgroundSdifference of threeSframeSdifference,then according to Mean-shift algorithm for tracking of moving targets.Finally,the moving objectsStrajectoriesSorthogonal decomposition,According to theStrajectory of theSX,SYSaxis component curve Sto identify theSwander behaviorS.TheSexperimentSresultsSshowSthatSthisSmethodScanSjudgeSseveral typicalShoveringSbehaviorSaccurately andSinSreal-time.alsoScan detectSother complexSwandering behavior,ShasSbetter real-time performance and veracity.