一种行人跟踪遮挡处理方法

An Algorithm for Occlusion Handling in Pedestrian Tracking

针对视频监控中目标遮挡问题,提出一种基于加权亮度直方图的行人跟踪遮挡处理方法。首先,利用卡尔曼滤波预测功能得到目标预测中心位置坐标并依此得出遮挡类型,同时预测遮挡时间。在遮挡前期,结合目标自身或目标间相对运动状态,对不同位置的像素亮度值赋以不同权值;在遮挡后期,反转权值计算公式,得到新的加权值。通过计算加权后的直方图之间的相似度进行遮挡跟踪。实验结果证明,加权直方图的方法可以有效降低遮挡带来的干扰,同时合理利用有效亮度信息,提高了跟踪系统的连续性和可靠性。

To solve the problem of occlusion in the intelligent video surveillance,an algorithm for occlusion handling in pedestrian tracking based on weighted luminance histogram is proposed in this paper. Firstly,we get the center coordinates of the objects that failed in matching through the prediction function of Kalman filter,then we decide which kinds of occlusion they are,and get the length of occlusion time. At the early stage of occlusion,different weighted factors are set to the pixels reference to the kinds of occlusion and the motion state of objects or the relative movements between the objects; at the later stage of occlusion,new factors are got by using the inversion equation.We can match the objects via calculating the similarity of the weighted luminance histograms. The experiment results indicate that this algorithm can make most of the luminance information of the uncovered parts,and reduce the impact of the occlusion. It works well in dealing with the occlusion in tracking.