基于GMM-RBMCDA的实时监控多目标跟踪方法

摘要

To extract the foreground objects and track the motions of these moning objects in the video surveillance scenes, this paper presented a GMM and Rao-Blackwellized Monte Carlo data association (GMM-RBMCDA) based on real-time multi-object tracking methood. The algorithm subtracted background adaptively by modeling each pixel as a mixture of Gaussians. And it used the RBMCDA algorithm to deal with the problem of trajectory-crossing and the background noise. Then it realized the real-time multi-object tracking by setting the parameters of objects appearance and disappearance. The experimental results show that the presented method can not only obtain the number of the objects in the video scenes,but also estimate the states of objects more accurately. It also proved the effectiveness of the algorithm.%为了有效提取视频监控场景中的前景目标信息并准确跟踪目标的状态,提出一种基于混合高斯模型和Rao-Blackwellized蒙特卡洛数据关联的视频多目标跟踪方法.该方法根据场景中像素点的特征信息,利用混合高斯模型进行建模,并对前景目标进行检测,使用Rao-Blackwellized蒙特卡洛数据关联算法来降低可能的目标交叉及杂波干扰带来的影响,通过设置目标存在和消失参数,实现了实时多目标跟踪.实验结果表明,该方法不仅能对场景中未知目标的个数进行有效估计,而且可以准确地跟踪目标的状态,取得了良好的实际效果.