基于改进SIFT算法的弹载电视制导技术研究  

Research on Missile-Borne TV Guidance Technology Based on Improved SIFT Algorithm

作　　者：刘桢 任梦洁 姜万里 LIU Zhen;REN Mengjie;JIANG Wanli(Laboratory of Guidance Control and Information Perception Technology of High Overload Projectiles,Army Artillery and Air Defense Officer Academy of PLA,Hefei 230031,China;School of Graduate management team,Army Artillery and Air Defense Officer Academy of PLA,Hefei 230031,China)

机构地区：[1]陆军炮兵防空兵学院高过载弹药制导控制与信息感知实验室,安徽合肥230031 [2]陆军炮兵防空兵学院研究生管理大队,安徽合肥230031

出　　处：《红外技术》2018年第3期280-288,共9页Infrared Technology

基　　金："十二五"装备预研基金重点项目(9140A05030213JB91013)

摘　　要：以图像匹配技术为代表的弹载电视制导技术具有信息直观的特点,作为非常优秀的图像匹配技术,SIFT算法受到了广泛的关注和深入的研究。针对传统SIFT算法实时性差的问题,本文提出了一种改进的SIFT算法。在提取特征点部分,通过Laplace算子找出图像边缘区域并进行Laplace加权处理,然后利用FAST特征点检测算法提取区域特征点;在生成特征点描述子部分,将传统的128维SIFT算子降为48维,利用改进的SIFT特征描述算子为特征点赋予方向和描述符使其具有旋转不变性;在特征点匹配部分,利用欧式距离提取匹配点对,并采用RANSAC算法提纯匹配点对,得到最优矩阵。实验结果表明改进的SIFT算法在目标旋转、尺度变化等条件下匹配效果良好,与传统SIFT算法相比具有很高的实时性,可以很好地实现图像实时匹配。The technology of missile-borne television guided by image matching technology has the characteristics of information visualization.As an effective image matching technology,the SIFT algorithm has received extensive research attention.Aiming at resolving the problem of the poor real-time performance of the traditional SIFT algorithm,an improved SIFT algorithm is proposed.In the extraction of feature points,the Laplace operator is used to find image edge regions and Laplace weighting is performed.In the generation of feature point descriptors,the traditional 128 dimensional SIFT operator is reduced to 48 dimensions,and the improved SIFT operator is adopted to assign directions and descriptors to feature points with rotation invariance.In the matching of feature points,the matching points are extracted by Euclide distance and are refined by the RANSAC algorithm to obtain the optimal matrix.The results of experiments show that the improved SIFT algorithm provides good matching effect under the conditions of target rotation and scale change.It performs well in real-time and can realize real-time image matching in comparison with the traditional SIFT algorithm.

关 键 词：LAPLACE算子 Laplace加权 FAST 改进SIFT特征描述算子 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]