针对红外序列图像中各类目标及背景特征动态变化的特性,提出一种基于二值分类技术的多特征融合目标跟踪算法.分别根据灰度、纹理及梯度方向特征将图像分为背景与目标区域,并根据各类特征分类性能的差异,融合特征图像,通过重采样粒子滤波估计目标状态.实验结果表明,该算法对环境光照变化、局部遮挡等均具有较好的鲁棒性.%Due to the various types of dynamic changes of background and foreground characteristics during object tracking in infrared image sequences, this paper proposes an object tracking algorithm of multi-feature fusion based on binary classification. The scene is classified into object and background region based on characteristics such as intensity, texture and grad orientation. The likelihood map is combined with the weights corresponding to classification performance respectively. A re-sampling particle filter is employed to estimate the object state. Experimental results show that the proposed algorithm is robust to environmental illumination and partial occlusions.
展开▼