摘要
为提高视觉同步定位与建图算法的精度和鲁棒性,提出一种基于半直接法的双目视觉SLAM系统。该系统基于LDSO,其视觉前端同时提取直接和间接特征,在跟踪过程中通过联合优化光度误差和重投影误差估计相机位姿,在后端建图过程中加入双目约束条件提高系统的绝对精度。在公开数据集EuroC、KITTI上与ORB-SLAM3等开源算法进行对比实验,其结果表明,所提算法较好结合了直接法、间接法和双目相机的优点,其平均定位精度较LDSO提升了34.6%,具有更好的鲁棒性。
To improve the accuracy and robustness of visual simultaneous localization and mapping algorithms,a semi-direct stereo visual SLAM system was proposed.Based on LDSO,both direct and indirect features were extracted in the visual frontend,camera poses were estimated by jointly optimizing photometric and reprojection errors during tracking,and stereo constraints were incorporated in the backend mapping process to enhance absolute accuracy.Results of comparative experiments with open-source algorithms like ORB-SLAM3 on public datasets EuroC and KITTI demonstrate that the proposed approach effectively combines the advantages of direct,indirect,and stereo camera methods.It not only enhances the average localization accuracy by 34.6%compared to that of LDSO,but exhibits superior robustness.
作者
唐帅
钟小勇
TANG Shuai;ZHONG Xiao-yong(College of Science,Jiangxi University of Science and Technology,Ganzhou 341000,China)
出处
《计算机工程与设计》
北大核心
2025年第4期997-1004,共8页
Computer Engineering and Design
基金
国家自然科学基金项目(51665019)。
关键词
同步定位与建图
直接法
间接法
半直接法
联合优化
双目相机
鲁棒性
simultaneous localization and mapping
direct method
indirect method
semi-direct method
joint optimization
stereo camera
robustness
