基于粒子滤波的交互式多模型说话人跟踪方法

An IMM Particle Filtering Method for Speaker Tracking

本文提出一种基于采样交互的多模型粒子滤波方法,实现了对随意运动说话人的有效跟踪.该方法根据说话人跟踪问题的特点,用马尔可夫跳变系统描述说话人的动态特性,用粒子滤波方法估计说话人的位置.在说话人跟踪过程中,通过调整滤波粒子的采样区域,完成交互式多模型方法中系统状态的交互过程,这不仅实现了各子滤波器中粒子数目的任意设定,避免了模型转换过程中的性能退化现象,而且取消了对模型后验概率密度函数的高斯分布假定,增强了说话人跟踪系统的鲁棒性.计算机仿真实验结果验证了本文方法的有效性.

A new interacting multiple model（IMM） algorithm based on particle filter is proposed to track a randomly moving speaker.Based on the characteristic of speaker tracking problem,the proposed method represents the dynamic model with Markov jump system and filtering the system state with particle filter.The interacting process is accomplished by properly selecting the sampling region.Thus,not only the number of particles in each mode can be controlled so that the degeneracy problem around mode transition is avoided,but also the Gaussian assumption of posteriori probability density function of the state is cancelled.Simulation results show the validity of the proposed method.