基于正则化观测矢量的H无穷粒子滤波红外目标跟踪方法

Infrared target tracking method based on regularized observation vector H infinity particle filter

提出了一种新颖和鲁棒的红外图像序列中的目标跟踪方法。由于H无穷滤波器在系统噪声源不能确定或是未知的情况下具有较好的预测性能,所以以其估计得到的预测信息来分配粒子滤波算法的粒子。为解决粒子滤波的"采样枯竭"问题,正则化了H无穷粒子滤波器的观测矢量。同时,通过计算每个目标的亮度和局部标准差分布构成级联核的目标模型,以用于计算粒子集中各个粒子的加权值。对于目标的尺寸和表观信息变化的情况,以目标区域像素灰度值零阶矩的函数来调整跟踪窗口的大小,模型更新则通过更新目标模型的每个量化阶来实现。实验结果证明了所提出的红外图像目标跟踪方法是有效的,并且优于所比较的算法。

A novel and robust approach for target tracking in infrared image sequence is proposed in this paper. Because H infinity filter has superior performance in prediction when the system has uncertain or unknown noise sources, the particles of the particle filter are allocated based on the prediction information estimated from H infinity filter. To overcome the sample impoverishment problem of particle filter, the observation vector of H infinity filter is regularized. At the same time, intensity and local standard deviation distributions of each target are computed to construct the cascaded kernel target model which is used to compute the weight of each particle in the particle set. In cases when the target scale and appearance change, tracking window size is set according to a function of the zero moment of the pixel gray values in the target region and update the target model for each bin. The experiments show the proposed infrared target tracking method is effective and superior to the compared algorithms.