基于形状模板匹配的前视红外目标检测方法  

FLIR Target Detection Algorithm Based on Shape Template Matching

作　　者：高晶[1] 孙继银[1] 吴昆[1] 刘婧[2] 

机构地区：[1]第二炮兵工程大学四系 [2]96411部队

出　　处：《北京工业大学学报》2012年第9期1359-1365,共7页Journal of Beijing University of Technology

基　　金：国家自然科学基金资助项目(61003148)

摘　　要：针对前视红外复杂地面固定目标无直接可用基准图、背景干扰严重、目标与背景灰度差异小、不利于目标识别等问题,提出了一种基于形状模板的目标识别方法.首先,在构建高斯多尺度空间的基础上,设计分层多阈值算法,检测感兴趣区域;其次,引入模糊集理论,提取形状特征,分离目标与背景;最后,用改进的Hausdorff距离算法进行精匹配,确定目标.实验结果表明,该算法匹配率与改进的Hausdorff距离算法相比提高了近20%,算法花费时间缩短了2/3;与Nprod算法相比匹配率提高了近30%,时间缩短了1/2,在密度为0.3的椒盐噪声下,匹配率仍能达到70%以上.对于复杂背景下的前视红外固定目标,该方法具有匹配率高、速度快、精度高等优点.For the forward looking infra-red （FLIR） image of complex ground fixed target without available base image, it was difficult to recognize the target due to the serious background clutter and small intensity differences of gray scale between target and background. A target recognition algorithm based on shape template matching was proposed. First, hierarchical multiple threshold algorithm based on constructing Gaussian multi-scale space was designed to test region of interest; Second, to extract shape feature and separate the target and background, fuzzy set theory was introduced; Finally, the modified Hausdorff distance algorithm was used for the precise matching, determining target. Experimental results show that comparing to modified Hausdorff distance algorithm, the matching probability of the proposed algorithm increases nearly 20% , taking time is shortened by 2/3, and comparing to Nprod algorithm, the matching probability increases nearly 30% , taking time is shortened by 1/2, under the pepper noise of 0.3 density, the matching probability is still able to reach 70% more. For recognising FLIR target on complex ground, this method has better performance on matching probability, computing speed and recognition precision.

关 键 词：尺度空间 分层多阈值 感兴趣区域 模糊集 目标识别 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]