基于双目视觉测量的特征匹配算法研究  

Research on Feature Matching Algorithm Based on Binocular Vision Measurement

作　　者：张静[1] 聂尚卿 张昊[1] 赵青伟 ZHANG Jing;NIE Shang-qing;ZHANG Hao;ZHAO Qing-wei(Energy and Power Engineering College,Lanzhou University of Technology,Lanzhou 730050,China)

机构地区：[1]兰州理工大学能源与动力工程学院,甘肃兰州730050

出　　处：《液压气动与密封》2025年第2期18-25,共8页Hydraulics Pneumatics & Seals

基　　金：国家自然科学基金(52165006)。

摘　　要：为了解决在接触式测量方法下液压元件表面产生磨损以及效率低下的问题,基于双目视觉非接触式测量技术,针对特征匹配过程中AKAZE算法匹配速度慢,RANSAC误匹配剔除算法模型的不稳定性以及冗余性问题提出一种改进方法,将Hessian矩阵特征点检测与M-LDB描述符相结合使用以提升匹配速度,并通过优化RANSAC算法迭代流程提升匹配精度。最后使用两种标准数据集对改进前后的两种算法进行实验验证,结果表明,改进算法在提升了匹配精度的同时也保证了匹配效率,具有良好的鲁棒性。In order to solve the problems of wear and low efficiency on the surface of hydraulic components under contact measurement methods,this paper proposes an improved method based on binocular vision non-contact measurement technology to address the slow matching speed of AKAZE algorithm,instability and redundancy of RANSAC mismatching removal algorithm model in feature matching process.The Hessian matrix feature point detection is combined with M-LDB descriptor to improve matching speed,and the matching accuracy is improved by optimizing the RANSAC algorithm iteration process.Finally,two standard datasets were used to experimentally validate the two algorithms before and after improvement.The results showed that the improved algorithm not only improved matching accuracy but also ensured matching efficiency,with good robustness.

关 键 词：特征匹配 AKAZE RANSAC 误匹配 描述符 

分 类 号：TH137[机械工程—机械制造及自动化] TP391.41[自动化与计算机技术—计算机应用技术]