粒子滤波算法在非线性目标跟踪系统中的应用

提出了一种基于贝叶斯理论及蒙特卡罗仿真的粒子滤波算法.该算法通过非参数化的蒙特卡罗(Monte Carlo)模拟方法来实现递推贝叶斯滤波,适用于任何能用状态空间模型以及传统的卡尔曼滤波表示的非线性系统,精度可以逼近最优估计.给出了算法的理论依据及整个跟踪过程的框架,并通过仿真试验对算法进行了验证.与传统的目标跟踪算法相比,本算法不仅能实现对目标的稳定、准确跟踪,将跟踪精度提高到90%以上,并且,当受到严重遮挡而发生目标丢失时,该算法仍然能够在10帧内重新捕获目标.实验结果证明,算法对于部分遮挡等复杂的非线性、非高斯情况具有良好的跟踪性能.

Nonlinear object tracking using particle filter

A particle filter algorithm based on Bayesian theory and Monte-Carlo simulation is presented. The algorithm adopts non-parameter Monte-Carlo simulation to realize Bayesian filter, and suits to any non-linear system represented by state space and conventional Kalman filter. The basic theories and the whole frame of the tracking system are given. Compared with the conventional tracking algorithms, the particle filter algorithm can track the object accurately and steadily, and the accuracy can be improved to more than 90%. Furthermore, when the tracker misses the object because of the severe occlusion, the algorithm can re-catch the object within 10 frames. The experiment results show that the algorithm can deal with the complicated nonlinear tracking problems well such as partial occlusion.