一种新的基于粒子滤波的多模型跟踪算法

粒子滤波技术通过非参数化的蒙特卡罗模拟方法实现递推贝叶斯滤波,适用于非线性目标运动模型、非线性传感器测量模型和非高斯噪声的目标跟踪。但需已知目标和量测模型,而实际情况往往难以满足此条件。交互多模型算法(IMM)依据各模型对目标前一时刻状态估计的方差,确定各模型在当前时刻状态下存在的概率,利用各模型对目标状态估计的加权和,确定目标的状态。本文采用粒子滤波代替IMM算法中各模型的Kalman滤波,将粒子滤波与IMM的优点相结合。同时,采用UKF(UnscentedKalmanFilter)产生粒子,由于考虑了当前量测,使得粒子的分布更加接近后验概率分布,用较少的粒子就可以逼近目标的真实状态。仿真实验结果表明,本算法可用于标准IMM算法无法实现跟踪的复杂情形,而且使用的粒子数目仅是同类算法的二十分之一。

A Novel Multiple Model Tracking Algorithm Based on Particle Filter

Particle filter realizes recursive Bayesian filter through non-parameter Monte Carlo technique, applying to target tracking in a setting where nonlinear target motions, non-Gaussian densities, or non-linear measurement to target couplings are involved. Unfortunately, it only tracks a trajectory with a known model at a given time, but for real world application, trajectory is always random in nature and may follow more than one model. Interactive Multiple Model (IMM) filtering mixes the state vector from different models using model probabilities that are based on the state covariance at a previous time from different models. In this paper, replacing Kalman filter with particle filter in each model of IMM algorithm integrates the advantages of particle filter with the ones of IMM. Meanwhile utilizing UKF (Unscented Kalman Filter) creates particles, thus making the distribution of particles closer to posterior probability density, and using a smaller number of particles to approximate to the true state of the target due to considering current measurements. The results of simulation show that the proposed algorithm can be used where standard IMM algorithm is impossible to apply, and moreover, that the number of particles used in the proposed algorithm is only a twentieth of ones in those available algorithms.