基于箱粒子滤波的多目标跟踪算法研究

发布时间：2018-12-14 01:50

【摘要】：目标跟踪由于应用广泛,受到了专家学者的普遍关注。在实际的跟踪场景中,感兴趣的目标往往不止一个,随着运动目标的出现和消失,目标的数目也是实时变化的,相应的多目标跟踪技术取得了巨大的发展。箱粒子滤波是近年来新提出的一种广义的粒子滤波方法,具有所需粒子数目少,计算复杂度低,计算效率高等优点。本文在箱粒子滤波基础上,对多目标跟踪方法进行了深入研究。介绍了箱粒子滤波的理论基础。箱粒子滤波本质上是广义的粒子滤波算法,它将区间分析这一数学工具与传统的蒙特卡洛算法相结合,用箱粒子代替了最大误差已知的点粒子,是一种处理非精确量测的方法。跟传统的粒子滤波算法相比箱粒子滤波算法体现出了良好的性能,在保持跟踪精度的前提下,所用粒子数目少,减少了算法的计算量,节省了运算时间,极大的提高了运算效率。本文在箱粒子和随机集的基础上,提出了一种新的多目标跟踪方法,箱粒子势概率假设密度滤波方法(BP-CPHD)。该算法保持了箱粒子滤波算法优点,又结合了CPHD滤波的优势。与传统的粒子CPHD算法相比,它的计算复杂度低,运算效率高。与基于箱粒子的概率假设密度(BP-PHD)算法相比,不需要对目标数目的分布做出符合泊松分布的假设,较好的解决了滤波器对杂波和漏检的敏感问题。通过递推目标数目的势分布,对目标数目做出了偏差更小的估计,从而提高了跟踪效果。在机动目标跟踪问题中,结合提出的基于箱粒子的势概率假设密度滤波(BPCPHD)算法和交互多模型算法,提出了交互多模型的箱粒子势概率假设密度滤波(IMM-BP-CPHD),该算法继承了箱粒子势概率假设密度滤波算法的优点,同时又能对多机动目标进行有效的跟踪,通过仿真实验,将该算法与区间量测下的交互多模型粒子势概率假设密度算法进行对比,体现了所提算法运行速度快等优点。
[Abstract]:Because of its wide application, target tracking has been paid more and more attention by experts and scholars. In the actual tracking scene, there is always more than one object of interest. With the appearance and disappearance of moving targets, the number of targets also changes in real time, and the corresponding multi-target tracking technology has made great progress. Box particle filter is a new generalized particle filter method proposed in recent years. It has the advantages of small number of particles, low computational complexity and high computational efficiency. On the basis of box particle filter, the multi-target tracking method is studied in this paper. The theoretical basis of box particle filter is introduced. Box particle filter is a generalized particle filter algorithm, which combines interval analysis, a mathematical tool, with the traditional Monte Carlo algorithm, and uses box particles instead of point particles with known maximum error. It is a method of dealing with imprecise measurement. Compared with the traditional particle filter algorithm, the box particle filter algorithm has a good performance. On the premise of keeping tracking accuracy, the number of particles used is less, the calculation amount of the algorithm is reduced, and the computation time is saved. The operation efficiency is greatly improved. On the basis of box particle and random set, a new multi-target tracking method, BP-CPHD (Particle potential probability assumption density filter), is proposed in this paper. The algorithm preserves the advantages of box particle filter and combines the advantages of CPHD filter. Compared with the traditional particle CPHD algorithm, it has low computational complexity and high computational efficiency. Compared with the probability assumption density (BP-PHD) algorithm based on box particle, it is not necessary to make the assumption that the distribution of target number accords with Poisson distribution, and the sensitivity of filter to clutter and miss detection is well solved. By recursive potential distribution of the number of targets, the deviation of the number of targets is estimated to be smaller, thus the tracking effect is improved. In the maneuvering target tracking problem, combining the (BPCPHD) algorithm based on the potential probability assumption density filter proposed by the box particle and the interactive multiple model algorithm, the paper proposes the box particle potential probability assumption density filter (IMM-BP-CPHD) based on the interactive multiple model. The algorithm not only inherits the advantages of the probability assumption density filter algorithm of box particle potential, but also can effectively track multiple maneuvering targets. The algorithm is compared with the interactive multi-model particle potential probability assumption density algorithm under interval measurement, which shows the advantages of the proposed algorithm, such as fast running speed and so on.
【学位授予单位】：西安电子科技大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：TN713

【参考文献】