论文部分内容阅读
由于目标数量的变化,观测数据的岐义性和目标间的遮挡,多目标视觉跟踪问题面临多种困难.基于目标分布的有限t分布混合模型提出了一种混合t分布粒子滤波器以实现多目标跟踪.在算法中,每个被跟踪目标指派一个独立的粒子滤波器,显式处理当新目标出现在场景中时对应粒子滤波器的初始化,当被跟踪目标消失时,对应粒子滤波器的删除.混合t分布粒子滤波器算法不仅能够跟踪多种类型的多目标,还能够持续跟踪遮挡消除之后的多目标.为了展现混合t分布粒子滤波器的跟踪性能,完成了基于颜色分布的跟踪多种不同颜色和相同颜色的多目标实验,对比了混合t分布粒子滤波器,混合粒子滤波器以及Boosted粒子滤波器的跟踪性能.实验结果表明:文中算法不仅能够跟踪数量可变的多目标,进行实时计算,而且具有更好的鲁棒性.“,”Abstract:Multiple objects visual track exhibits a number of difficulties due to the variable number of objects,the ambiguity of the observations and the presence of partial occlusions.Since the representing target distributed as finite t distribution mixture models,a mixtures of t distribution particle filters (MTPF) was introduced to track multiple objects.In this paper,each tracked object was assigned a separate particle filter to handle the instantiation and removal of filters when new objects entered the scene or previously tracked objects were removed.The proposed filter can not only track multiple objects of different types,but also continue to track objects through occlusion situations.To present the tracking performance of the MTPF algorithm,experiments were performed using color-based tracking of multiple objects with different and identical colors,and comparison experiments between mixtures of t distribution particle filters,mixtures of particle filters and the boosted particle filters were conducted.Experimental results demonstrate the proposed tracking algorithm not only track a variable number of targets with the real-time computation,but also perform the more robustness.