一种仿射不变的前视红外目标识别方法  

An Affine Invariant Method of Forward Looking Infra-Red Target Recognition

作　　者：赵爱罡[1] 王宏力[1] 杨小冈[1] 陆敬辉[1] 王建永 崔祥祥[1] 

机构地区：[1]第二炮兵工程大学303教研室,陕西西安710025

出　　处：《激光与光电子学进展》2015年第7期171-178,共8页Laser & Optoelectronics Progress

基　　金：国家自然科学基金(61203189);总装预研基金(9140A01060411JB4701)

摘　　要：前视红外目标识别是精确制导武器的关键技术。实际应用中,模板和实时图存在尺度和角度的差异,易导致误匹配。为克服上述问题,提出了一种基于边缘仿射不变坐标的前视红外目标识别算法。以边缘曲线的区域质心为基准点建立局部坐标系,在此坐标系下,边缘点的坐标值具有仿射不变;将边缘的坐标集合作为描述子,利用平均hausdorff距离度量边缘的相似性;最后结合边缘的相对位置剔除误匹配,实现前视红外目标的稳健识别。通过三组实验,与基于灰度模板和基于边缘模板的识别算法相比,识别率和稳健性均有所提高。Forward looking infrared target recognition is key technology in precision-guided weapon. In practical application, it is likely to cause wrong matching because of discrepancy of scale and angle between template and real images. In order to solve problem mentioned above, A forward looking infrared target recognition algorithm based on edge＇s affine invariant coordinate. Establishing coordinate system based on regional centroids of curvic edge. Coordinates of edge points is affine invariant in the established coordinate system. Taking the set of edge points＇ coordinates as edge descriptor and similarity of edge is measured by average hausdorff distance. Wrong matchings are excluded with the help of relative locations among edges and forwarding looking infrared target robust recognition is implemented. Compared with two recognition algorithms based on gray template and edge template, recognition rate and robustness of proposed algorithm are all improved through three videos.

关 键 词：图像处理 仿射不变 目标识别 HAUSDORFF距离 轮廓曲线 红外图像 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]