基于模型动态切换的运动目标实时跟踪

针对视频监控系统中运动目标的跟踪问题,提出了一种基于模型动态切换的实时跟踪方法.在运动目标分割之后,跟踪系统有效判定运动目标的遮挡状态,对未遮挡的运动对象采用基于区域的跟踪模型,对于相互重叠的运动对象采用基于SIFT特征的窄基线图像匹配模型.基于区域的跟踪模型采用简单的目标区域特征以及运动预测属性,实现快速地跟踪.基于SIFT特征的图像匹配模型利用被跟踪目标在相邻图像帧之间很小的尺度和外形变化以及基于目标区域位置预测出的有限运动范围,实现快速的窄基线小范围SIFT特征匹配和跟踪.实验结果表明,该方法具有较强的鲁棒性,能有效实现复杂遮挡场景下的多目标实时跟踪.

Real-time moving target tracking based on dynamically switching models

Aiming at moving target tracking in video surveillance systems,a real-time tracking method based on dynamically switching models is proposed.After moving target segmentation,the tracking system effectively estimates the occlusion state of moving objects.A region-based tracking model is deployed for non-occluded moving objects,and a narrow-scale image matching model based on SIFT(scale invariant feature transform) features is deployed for occluded moving objects.The region-based tracking model uses simple target region features and motion prediction attributes to implement rapid tracking.Because the tracked targets only have very small scale changes and limited range on target region position prediction between neighboring image frames,the image matching model based on SIFT features can also implement rapid SIFT features matching and tracking.Experimental results show that the method is robust and can effectively achieve realtime multi-target tracking in complex occlusion scenes.