基于BEBLID描述子与SuperGlue补偿匹配改进的ORB-SLAM3算法  

Improved ORB-SLAM3 algorithm based on BEBLID descriptor and SuperGlue compensation matching

作　　者：赵义武 王星朝[1] 黄丹丹 刘智 郭星辰[4] 侯舒帆 ZHAO Yiwu;WANG Xingzhao;HUANG Dandan;LIU Zhi;GUO Xingchen;HOU Shufan(School of Computer Science and Technology,Changchun University of Science and Technology,Changchun 130022,China;School of Electronic Information Engineering,Changchun University of Science and Technology,Changchun 130022,China;Institute of Space Photo-Electronic Technology,Changchun University of Science and Technology,Changchun 130022,China;School of Opto-electronic Engineering,Changchun University of Science and Technology,Changchun 130022,China)

机构地区：[1]长春理工大学计算机科学技术学院,长春130022 [2]长春理工大学电子信息工程学院,长春130022 [3]长春理工大学空间光电技术研究所,长春130022 [4]长春理工大学光电工程学院,长春130022

出　　处：《中国惯性技术学报》2025年第3期249-256,共8页Journal of Chinese Inertial Technology

基　　金：国家重大科研仪器研制项目(62127813);吉林省科技厅重点研发项目(20230201071GX)。

摘　　要：为了提高同时定位与地图构建(SLAM)系统的精确性和稳定性,在ORB-SLAM3的基础上提出了一种基于BEBLID描述子与SuperGlue补偿匹配的视觉SLAM算法。首先,对图像数据进行特征提取,计算BEBLID描述子,并利用这种二进制描述子进行特征匹配;然后,针对匹配效果不稳定的问题,采用SuperGlue算法对具备映射关系但未成功匹配的特征点对补偿匹配,以提高跟踪的稳定性。实验结果表明,相比于ORB-SLAM3算法,使用BEBLID描述子进行特征匹配能够平均降低16.08%的绝对轨迹误差;使用SuperGlue进行补偿匹配能够提高算法跟踪效果的稳定性。综合应用BEBLID描述子和SuperGlue补偿匹配策略,可以显著提高SLAM系统的鲁棒性。To enhance the accuracy and stability of the simultaneous localization and mapping(SLAM) system,building on ORB-SLAM3,a visual SLAM algorithm is proposed incorporating BEBLID descriptors and SuperGlue matching compensation.First,extracting feature points from the image,computing the BEBLID descriptor,and utilizing this binary descriptor for feature matching.Then,to address the issue of unstable matching results,the SuperGlue algorithm is used to compensate for the matching of feature point pairs that have mapping relationships but have not been successfully matched to enhance tracking stability.Experimental results demonstrate that compared to the ORB-SLAM3 algorithm,using BEBLID descriptor for feature matching can improve the average absolute trajectory error accuracy by 16.08%.Additionally,by utilizing SuperGlue for matching compensation,the tracking performance of the algorithm becomes more stable.The comprehensive application of BEBLID descriptor and SuperGlue compensation matching strategy can significantly improve the robustness of visual SLAM system.

关 键 词：同时定位与地图构建 特征匹配 BEBLID描述子 SuperGlue算法 补偿匹配 

分 类 号：TP242[自动化与计算机技术—检测技术与自动化装置]