基于梯度方向-亮度联合空间的红外人体跟踪

Human tracking based on gradient orientation-intensity joint space in infrared image sequences

针对红外图像中人体成像的特点,提出一种新颖的红外人体跟踪算法。为克服红外图像中人体目标描述信息不足的缺点,该方法首先在梯度方向-亮度二维联合空间中构建人体目标的特征直方图,然后给出了一种最优直方图级数的选择准则对最具鉴别性能的级数进行选择;进而将上述表达模型与粒子滤波相融合,设计了粒子滤波框架下的人体跟踪算法。不同场景中的人体跟踪结果表明,与通用的跟踪算法相比,本文提出的算法具有更高的鲁棒性和稳健性。

According to the imaging features of human body in infrared system, a novel human tracking approach in infrared image sequences is proposed. The representing histogram of the human body in a joint space of gradient orientation and pixel intensity is constructed, so as to hurdle the disadvantage of insufficient information when only intensity feature is used. Also, an evaluation criterion is presented to get the optimal bins of the histogram. The above mentioned human representing model is embedded in a particle filter frame, and the human tracking algorithm is designed under this frame. Experimental results under different scenarios demonstrate that the proposed method is more robust and stable than the classical tracking method.