一种鲁棒的多目标自动跟踪算法

A robust automatic multi-target tracking algorithm in complex environments

针对复杂的自然环境,提出了一种能自动初始化目标模型,并能处理目标遮掩问题的多目标实时跟踪算法。该算法在精度和效率方面改进了Mean Shift算法,并结合扩展卡尔曼滤波器对目标进行运动建模,实现了对多目标的实时稳健跟踪。此外,为了自动初始化待跟踪目标模型,提出了一种三层目标链交互结构。该结构能够有效的去除由摄像机抖动以及背景噪声产生的伪目标。在复杂的自然环境下对算法进行了大量的多目标跟踪实验,验证了算法的实时有效性。

A real-time multi-target tracking method is presented in this paper, which can automatically initialise the models of targets and can solve the problems of occlusions ,especially dealing with the tracking under complex natural environment. This method improves the classic Mean Shift method in aspects of precision and efficience, and models the motion of targets by the Extended Kalman Fiter, to stably track multiple targets. Moreover, to initialize the target models automatically, a Tri-MotionRegion-List Interaction structure is proposed, which can efficiently erase out the false targets brought by the camera dithering and background noises. A lot of experiment resuits under complex natural environments show that this method is efficient and performing in real-time.