基于无味粒子滤波的动态场景下高机动目标跟踪

Maneuverable target tracking in dynamic scene based on unscented particle filter

针对传统粒子滤波的目标跟踪算法存在粒子退化问题,提出了基于无味粒子滤波(UPF)的目标跟踪算法。为了将当前观测信息融入,采用无味卡尔曼滤波(UKF)生成粒子滤波的提议分布,以改善滤波效果。针对目标在机动过程中引起的视觉形变以及背景的变化,又采用了颜色直方图作为目标的颜色分布模型,并与UPF相融合。仿真结果表明,该算法对动态场景下的高机动目标有较好的跟踪效果。

Target tracking based on the unscented particle filter（UPF） is proposed to solve the problem of particles degradation in the general particle filter algorithm.To solve the problem that the transition prior does not into account the current observation in the general particle filter,instead of using transition prior as proposal distribution,the unscented Kalman filter（UKF） is used to generate the proposal distribution,which improves the filtering effect.According to the visual deformation and scene change caused by target maneuvering,the HSV color histogram is adopted as targets′ color distribution model and fused with the unscented particle filter（UPF）.The simulation results show that the proposed algorithm can effectively track maneuvering target in dynamic scene with much better performance.