摘要
针对弱纹理场景下点特征提取不足影响同步定位与建图(SLAM)算法定位精度的问题,提出了一种室内环境下点线特征综合的RGB-D视觉里程计(VO)算法。通过跟踪深度信息计算垂直主导方向,基于曼哈顿假设使用线特征来加权搜索两个水平自由度,提取并优化曼哈顿坐标系;综合场景的结构规律,与点线特征的重投影误差进行联合优化,同时在位姿估计和局部地图优化中对残差引入自适应权重,提高位姿估计精度。实验结果表明,所提算法在ICL-NUIM数据集中的绝对轨迹均方根误差相比于ORB-SLAM2和MSC-VO分别平均减少62.93%、37.04%,与Planar-SLAM和Manhattan-SLAM精度相当;在TUM数据集中相比于ORB-SLAM2、Planar-SLAM、MSC-VO和Manhattan-SLAM,绝对轨迹均方根误差分别平均减小21.43%、54.40%、35.08%和26.94%;在TAMU数据集中相比于ORB-SLAM2,回环漂移平均减小43.34%。
Aiming at the problem that the localization accuracy of simultaneous localization and mapping(SLAM)algorithm is affected by the insufficient extraction of feature points in weak texture scenes,a RGB-D visual odometry algorithm integrating point-line features is proposed.The vertical dominant direction is calculated by tracking the depth information.Based on the Manhattan assumption,the line features are used to weighted search two horizontal degrees of freedom,and the Manhattan frame is extracted and optimized.The structural information of the scene is integrated with the reprojection of the point and line feature errors for joint optimization,and at the same time,adaptive weights are introduced to the residuals in the pose estimation and local map optimization to improve the pose estimation accuracy.The experimental results show that,compared with ORB-SLAM2 and MSC-VO,the absolute trajectory root mean square error of the proposed algorithm is reduced by 62.93%and 37.04%on average respectively,and is comparable to Planar-SLAM and Manhattan-SLAM in ICL-NUIM dataset;compared with ORB-SLAM2,Planar-SLAM,MSC-VO and Manhattan-SLAM,the absolute trajectory root mean square error of the proposed method is reduced by 21.43%,54.40%,35.08%and 26.94%on average in TUM dataset,respectively;and compared with ORB-SLAM2,the loop drift is reduced by 43.34%on average in TAMU dataset.
作者
程向红
刘路辉
唐兴邦
CHENG Xianghong;LIU Luhui;TANG Xingbang(Key Laboratory of Micro-inertial Instrument and Advanced Navigation Technology,Ministry of Education,Southeast University,Nanjing 210096,China;School of Instrument Science&Engineering,Southeast University,Nanjing 210096,China)
出处
《中国惯性技术学报》
EI
CSCD
北大核心
2024年第6期579-585,共7页
Journal of Chinese Inertial Technology
基金
国家科学自然基金(62273091)。
关键词
点线特征
曼哈顿假设
视觉里程计
point and line features
Manhattan assumption
visual odometry
