融合梯度特征的红外目标跟踪  

The Infrared Target Tracking by Fusion of Gradient Character

作　　者：余博[1] 郭雷[1] 赵天云[1] 钱晓亮[1] 左蔚[2] 

机构地区：[1]西北工业大学自动化学院,西安710129 [2]国家开发银行信息科技局,北京100044

出　　处：《西安工业大学学报》2012年第5期355-360,共6页Journal of Xi’an Technological University

基　　金：航空科学基金(20080153002)

摘　　要：Mean shift跟踪算法普遍采用单一的灰度空间进行红外目标跟踪,直方图所含信息量少,容易受目标和背景灰度变化的影响.文中提出一种融合梯度特征的红外图像Mean shift跟踪算法,该算法对边缘和结构特征的梯度值进行量化,建立梯度特征模型;利用Bhatta-charyya系数分别计算梯度特征和灰度特征的特征相似度;设置置信度系数,并利用置信度系数将梯度特征相似度和灰度特征相似度进行融合,得到综合相似度;针对综合相似度推导Mean shift迭代方程,通过迭代运算逐步逼近目标实现跟踪;利用灰度特征相似度和梯度特征相似度信息并结合置信度系数设计模型更新准则以提高跟踪鲁棒性.文中算法能够适应红外目标跟踪中目标与背景的变化,鲁棒性强且跟踪准确度高,仿真实验表明该算法较普通Mean shift算法性能有较大提高,跟踪精度也有所改善.The mean shift algorithm of infrared-target tracking usually works in the single gray space. Because of the little information of histogram, the algorithm is liable to be affected by the change of targets and background. In the paper, the infrared-target tracking algorithm by fusion of gradient character is proposed. Firstly, a target model and candidate model of gradient character are built, and the gradient similarity between the target model and the candidate model is calculated with Bhattacharyya coefficient. Secondly, the gradient similarity and the gray similarity are fused by the Log likelihood ratio. The formula of mean shift iterative is obtained, and then the target can be tracked by iterative operation. Finally, the updating standard of infrared-image is designed by the gray similarity, gradient similarity and the Log likelihood ratio. This algorithm can adapt to the change of target and background in the infrared-image tracking. The algorithm is robust and tracks accurately. Experiments show that this algorithm is superior in performance and precision to the ordinary mean shift algorithm.

关 键 词：梯度特征 Mean SHIFT 目标跟踪 红外图像 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]