基于QK-GMPHD的数目可变视频多目标跟踪方法

针对非线性系统模型多目标跟踪问题，提出了一种基于求积分卡尔曼．概率假设密度滤波器。利用高斯混合概率假设密度滤波器对后验多目标状态的一阶矩进行估计，并通过求积分卡尔曼递推更新得到目标状态，以实现对目标数目时变的多目标进行跟踪。求积分卡尔曼滤波器使用统计线性回归的方法，通过一系列高斯-厄米特积分点来线性化非线性函数，无需求解非线性系统函数的雅克比矩阵，降低了线性化误差。将本文算法与全局最近邻数据关联方法相结合，解决非线性系统数目可变视频多目标跟踪问题。冰球视频运动员跟踪结果表明，本文方法能够解决新目标出现、目标合并、目标分裂以及目标消失问题，实现数目可变视频多目标跟踪。

QK-GMPHD based variable number of video targets tracking

Aiming at nonlinear system model in multi-target tracking, a new quadrature Kalman-probability hypothesis density filter is presented. The algorithm estimates first-order statistical moment of posterior multi-target states with Gaussian mixture probability hypothesis density （GMPHD） filter, and updating targets＇ states from the recursion of uadrature Kalman （QK）. Quadrature Kalman filter uses statistical linear regression method through a set of Gaussian-Hermite quadrature points to linearize nonlinear functions Jacobian matrix solving is unnecessary and the linearized error is reduced. The proposed algorithm combines with global nearest neighbor （GNN） data association method is presented for nonlinear system variable number of video targets tracking. Ice hockey players tracking results show that QK-GMPHD based multi-target visual tracking can track objects correctly when they appear, merge, split and disappear in the field of view of a camera, and effectively track varying number of targets.