基于背景差分检测和改进GM-PHD滤波器的多目标跟踪

摘要

Target label confusion and loss are usually caused by occlusion and detection missing in multiple object tracking process,which leads to failing tracking.Aiming at this problem,an improved tracking method based on Gaussian Mixture Probability Hypothesis Density(GM-PHD) filter is proposed.The binary image mapping and testing sets are got by Background Subtraction Detection(BSD),and the object appearance is detected by detector based on the appearance.The two testing sets got by background subtraction and appearance detector are fused.The improved GM-PHD filter is used to keep the object tracking trajectory so as to deal with some uncertainty in object tracking.Experimental results show that the tracking precision of the proposed method is superior to that of GM-PHD method,color appearance method and SMC-PHD method.%在多目标跟踪过程中,遮挡和漏检容易引起目标标签错乱和丢失,造成跟踪失败.针对该问题,提出一种基于混合高斯-概率假设密度(GM-PHD)滤波器的改进跟踪方法.使用背景差分检测获得二值图像映射和测量集,以外观为基础的探测器检测目标外观,将背景差分获得的测量集与外观检测器获得的测量集进行融合,利用改进的GM-PHD滤波器保持目标跟踪轨迹,并处理目标跟踪中的一些不确定性因素.实验结果表明,与GM-PHD方法、颜色外观方法和SMC-PHD方法相比,该方法能获得较好的跟踪精度.