基于局部显著边缘特征的红外图像目标轮廓跟踪算法  

Infrared Image Target Contour Tracking Algorithm based on Locally Significant Edge Features

作　　者：张正本 刘丹 ZHANG Zheng-ben;LIU Dan(School of Computer Science&Technology,Henan Institute of Technology,Xinxiang 453003,China;Xinxiang Manufacturing Engineering Center for IOT,Xinxiang 453003,China)

机构地区：[1]河南工学院计算机科学与技术学院,河南新乡453003 [2]新乡市制造业物联网应用工程技术研究中心,河南新乡453003

出　　处：《河南工学院学报》2020年第5期11-15,共5页Journal of Henan Institute of Technology

摘　　要：传统的红外图像轮廓跟踪算法跟踪精度较低。针对这一问题,提出了一种基于复杂背景的具有局部显著边缘特征的目标轮廓跟踪算法。该算法首先引入射影不变量,构造红外图像边缘位置之间的几何信息描述符,建立各目标轮廓特征数的直方图。利用巴氏系数度量特征间的几何相似性,建立目标轮廓周围邻域的边缘特征,搜索图像边缘具有显著特征的目标轮廓,形成特征描述向量,定义欧氏距离跟踪测量函数。利用该功能,对选定的目标轮廓进行初步跟踪。采用随机一致性检验算法消除伪跟踪特征点,得到目标轮廓的最佳跟踪值,从而在复杂背景下对红外图像进行目标轮廓跟踪。实验仿真表明,该算法具有较高的跟踪精度,有效地提高了红外图像分析的质量。Traditional infrared image contour tracking algorithms have low tracking accuracy.To solve this problem,a target contour tracking algorithm based on locally significant edge features in complex background is presented.The algorithm first introduces projective invariants,constructs geometric information descriptors between the edge locations of infrared images,and establishes histograms of the feature numbers of each target contour.The geometric similarity between features is measured by Bhattacharyya coefficients,and the edge features of the neighborhood around the target outline are established to search for the target outline with significant features on the edge of the image.The shape context operator is combined with the edge feature to form a feature description vector and define the Euclidean distance tracking measurement function.With th is function,the selected target contour is tracked preliminarily.The random consistency test algorithm is used to eliminate pseudo-tracking feature points and obtain the best tracking value for target contour,so that the infrared image can be tracked against a complex background.The experimental simulation shows that the algorithm has high tracking accuracy and effectively improves the quality of infrared image analysis.

关 键 词：边缘特征 红外图像 物体轮廓 跟踪 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]