首页|期刊导航|电子科技|基于结构垂直线支持的点线SLAM算法

基于结构垂直线支持的点线SLAM算法OA

Point Line SLAM Algorithm Based on Structural Vertical Line Support

中文摘要英文摘要

传统视觉同时定位与地图构建(Simultaneous Localization and Mapping,SLAM)方法使用点特征来估计摄像机运动轨迹并构建环境地图.针对地下停车场等特殊场景中的地面反射、光照变化等因素使传统点SLAM系统性能下降问题,文中提出了融合结构垂直线特征与点特征的VPL(Vertical-Point-Line)-SLAM算法,以此增强在特殊场景下位姿估计的鲁棒能力.改进直线检测模型ED-Lines(Enhanced Direct Lines),通过加入结构垂直线约束、添加额外的角度和长度标准剔除有噪声的线特征以及LBD(Line Band Discriptor)和光流的双策略匹配机制等方法获得高质量的直线特征,结合基于欧几里得距离的线重投影误差和结构方向误差构建更精确的线特征约束,以紧耦合方式构建点-线投影模型优化位姿估计.基于KITTI数据集以及自采集的数据集进行测试,实验结果表明,在两个数据集上,相较于原算法,所提算法的绝对平移误差均降低了约75%,增强了其在特殊场景下的鲁棒性.

The traditional visual SLAM(Simultaneous Localization and Mapping)method uses point features to estimate the camera's movement trajectory and construct an environmental map.In view of the problem that fac-tors such as ground reflection and illumination changes in special scenarios like underground parking lots cause the performance degradation of traditional point SLAM systems,this paper proposes a VPL(Vertical-Point-Line)-SLAM algorithm that fuses structural vertical line features and point features to enhance the robustness of pose estimation in special scenarios.The line detection model E-Lines(Enhanced Direct Lines)is improved.High-quality line features are obtained by adding structural vertical line constraints,adding additional angle and length standards to eliminate noisy line features,and the dual-strategy matching mechanism of LBD(Line Band Discrip-tor)and optical flow.More accurate line feature constraints are constructed by combining the line reprojection error based on Euclidean distance and the structural direction error.Construct a point-line projection model in a tightly coupled manner to optimize pose estimation.Tests are conducted on the KITTI dataset and self-collected datasets.The experimental results show that on both datasets,compared with the original algorithm,the absolute translation error of the algorithm proposed in this paper is reduced by approximately 75%,enhancing its robustness in special scenarios.

纪周;蔡洁;魏鸿扬

上海理工大学 光电信息与计算机工程学院,上海 200093上海理工大学 理学院,上海 200093上海理工大学 光电信息与计算机工程学院,上海 200093

信息技术与安全科学

同时定位与地图构建特殊场景SLAM系统点特征结构垂直线特征EDLinesLBDKITTI数据集

simultaneous localization and mappingspecial sceneSLAM systempoint featurevertical line characteristics of the structureEDLinesLBDKITTI dataset

《电子科技》 2026 (9)

58-66,9

国家自然科学基金(62273239)National Natural Science Foundation of China(62273239)

10.16180/j.cnki.issn1007-7820.2026.09.008

评论