传感技术学报2025，Vol.38Issue(7)：1199-1207,9.DOI:10.3969/j.issn.1004-1699.2025.07.008
具有结构特征辅助的RGB-D视觉里程计
RGB-D Visual Odometry with Structural Feature Assistance
摘要
Abstract
In indoor environments with sparse textures and uneven lighting,an RGB-D visual odometry with structural feature assistance is proposed to address the problems of feature degradation and localisation misalignment in traditional point-line visual odometers.First,to overcome the problem of poor robustness of line feature extraction due to uneven indoor illumination,the edge-fitting line feature ex-traction method is improved by introducing the pixel connection rule with horizontal and vertical constraints,so as to make the extracted line segments have better continuity and robustness.Then,the intersection relationship of the line features is utilised to construct a more stable node-line structural feature,to reduce the rotational cumulative error,and to improve the positioning accuracy.Finally,the fusion point-line BA optimisation is used to construct the minimum point-line reprojection error function to further improve the performance of the algorithm.The experimental comparison results on the TUM RGB-D dataset and the ICL-NUIM dataset with simulated illumination variations show that the proposed method can effectively improve the robustness of feature extraction,and make the position estimation more robust and accurate.关键词
视觉里程计/结构特征/线特征/位姿估计Key words
visual odometry/structural features/line features/position estimation分类
信息技术与安全科学引用本文复制引用
孙新柱,龚光强,陈孟元..具有结构特征辅助的RGB-D视觉里程计[J].传感技术学报,2025,38(7):1199-1207,9.基金项目
安徽省重点研究与开发计划项目(202304a05020073) （202304a05020073）
安徽省高校协同创新项目(GXXT-2021-050) （GXXT-2021-050）
安徽省高校杰出青年科研项目(2022AH020065) （2022AH020065）
安徽省学术和技术带头人后备人选科研活动经费择优资助(项目2022H292) （项目2022H292）
