论文部分内容阅读
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 target tracking method in infrared image sequence is proposed. Because the H-infinite filter has good prediction performance under the condition that the system noise source can not be determined or is unknown, the particle of the particle filter algorithm is allocated based on the estimated prediction information. In order to solve the “sampling depletion” problem of particle filter, the observational vector of H-infinite particle filter is regularized. At the same time, a target model of a cascade kernel is formed by calculating the brightness and local standard deviation distribution of each target for calculating the weighted values of each particle in the particle set. For the change of target size and apparent information, the size of the tracking window is adjusted as a function of the zero-order moment of the pixel gray value in the target area. The model updating is achieved by updating each quantization step of the target model. The experimental results show that the proposed method of tracking the infrared images is effective and superior to the algorithm under comparison.