多特征提取的红外和可见光目标跟踪方法  

A Method for Tracking Object in Infrared and Visible Image Based on Multiple Features

作　　者：赵高鹏[1] 薄煜明[1] 陈益[1] 

机构地区：[1]南京理工大学自动化学院,江苏南京210094

出　　处：《兵工学报》2011年第4期445-451,共7页Acta Armamentarii

摘　　要：针对复杂背景下采用单一图像特征跟踪稳定性差的问题,提出了一种基于多个特征提取的红外和可见光图像目标跟踪方法。该方法提取红外图像的灰度特征和可见光图像的颜色、纹理特征,给出了采用核函数量化的直方图描述的多特征目标模型,结合均值漂移跟踪框架,给出了一种根据相似性度量进行自适应融合定位目标的方法,并给出了一种选择性的目标模板模型更新方法。实验采用复杂地面环境下的多组图像序列,结果表明该方法的有效性,实现了复杂环境下的图像目标稳定跟踪。Aimed at poor stability of tracking object in image by using single feature,a new method was presented for tracking objects in infrared and visible images based on multiple features.It extracted the brightness features in infrared image,the color and texture features in visible image,established an multiple feature object model represented by a set of histogram quantized by kernel function.Then the object representation was combined with the mean shift scheme to realize image tracking,and these features were adaptively fused to obtain a more robust and accurate result according to the similarity measure of Bhattacharyya coefficient.Furthermore,a selective model update mechanism was presented to alleviate the model drift.Experiment results show that the proposed method is effective in complex environment.

关 键 词：信息处理技术 目标跟踪 多特征提取 均值漂移 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]