一种基于非线性尺度空间的ORB特征匹配算法研究  

Research on ORB Feature Matching Algorithm Based on Nonlinear Scale Space

作　　者：郭芮 许钢[1] 李若楠 赵春叶 江娟娟[1] GUO Rui;XU Gang;LI Ruonan;ZHAO Chunye;JIANG Juanjuan(Key Laboratory of Detection Technology and Energy￣saving device,Anhui Polytechnic University,Wuhu 241000,China)

机构地区：[1]安徽工程大学检测技术与节能装置安徽省重点实验室

出　　处：《贵州大学学报（自然科学版）》2020年第1期78-85,共8页Journal of Guizhou University:Natural Sciences

基　　金：安徽省重点实验室开放基金项目资助(2017070503B026-A01);安徽高校自然科学研究重点项目资助(KJ2018A0111)

摘　　要：ORB(Oriented FAST and rotated BRIEF)特征描述算法具有旋转不变性、匹配速度快的特点,但没有解决尺度不变性、误匹配率高的问题。针对此缺陷,提出一种改进的ORB特征点匹配算法,完成特征点的检测、匹配以及剔除误匹配。改进算法首先借鉴了A-KAZE基于非线性扩散滤波构建尺度空间的方法;其次利用ORB特征检测子在所构建的非线性尺度空间进行特征点的检测;再次对采集到的特征点生成特征描述子;最后在使用Hamming距离匹配的基础上再对其结果采用PROSAC算法剔除噪声点。实验结果表明,改进后的算法相较于原ORB算法,有效地解决了ORB算法不具备尺度不变性的问题,且匹配精度大幅提高,适用于尺度变化较大且实时性高的环境,具有较好的工程意义。The ORB(Oriented FAST and rotated BRIEF)feature description algorithm has the characteristics of rotation invariance,matching speed,but not solve the scale invariance,the problem of high error matching rate.Aiming at this defect,an improved ORB feature point matching algorithm is proposed to detect,match and eliminate mismatches.The improved algorithm first uses the method of A-KAZE based on nonlinear diffusion filtering to construct scale space.Secondly,the ORB feature detector is used to detect the feature points in the constructed nonlinear scale space.Then generate feature descriptors for the collected feature points.Finally,on the basis of hamming distance matching,PROSAC algorithm is adopted to eliminate noise points.The experimental results show that compared with the original ORB algorithm,the improved ORB algorithm effectively solves the problem that the ORB algorithm does not have scale invariance,and the matching accuracy is greatly improved.It is suitable for the environment with large scale variation and high real-time performance,and has good engineering significance.

关 键 词：改进ORB算法 A-KAZE 非线性扩散滤波 尺度不变性 特征点匹配 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]