线特征优化配置在室内RGB-D SLAM系统中的应用  

Application of optimal configuration-based line features in indoor RGB-D SLAM system

作　　者：夏琳琳[1] 崔家硕 宋梓维 胡源子 胡智棋 XIA Linlin;CUI Jiashuo;SONG Ziwei;HU Yuanzi;HU Zhiqi(School of Automation Engineering,Northeast Electric Power University,Jilin 132012,China)

机构地区：[1]东北电力大学自动化工程学院,吉林132012

出　　处：《中国惯性技术学报》2022年第6期760-767,776,共9页Journal of Chinese Inertial Technology

基　　金：吉林省科技厅自然科学基金项目(20220101240JC);吉林市科技局杰出青年人才培养专项(20190104128)。

摘　　要：运动估计是视觉同时定位与建图(SLAM)研究中的核心问题。从几何-物体特征基元的提取、匹配(关联)和参数化研究入手,结合因子图优化,构建室内RGB-DSLAM系统框架。将“线特征优化配置”策略引入至点-线-平面多观测约束的位姿估计,通过ICL-NUIM和TUM基准数据集实验,确定最优的线特征提取数量,提升系统在弱纹理区域的定位鲁棒性,并增强结果的可解释性。与ORB-SLAM2、PL-SLAM及SP-SLAM的对比实验表明:所开发的系统取得了最优的全局定位性能,相较SP-SLAM在ICL-NUIM数据集8个子序列、TUM数据集4个子序列上的平均轨迹估计精度分别提升了4.33%和21.40%。Motion estimation is the core problem of visual simultaneous localization and mapping(SLAM)studies. Starting from the extraction, matching(association) and parameterization of geometric-object feature primitives, combined with factor graph optimization, the indoor RGB-D SLAM system framework is constructed. The strategy of ’optimal configuration of line features’ has been applied to the point-line-plane constraints fused pose estimation, and the number of optimal line feature extraction is determined in terms of benchmark ICL-NUIM and TUM dataset tests. The localization robustness under weakly textured scenes, as well as the result interpretability, is therefore enhanced. The comparative tests on ORB-SLAM2, PL-SLAM and SP-SLAM illustrate that the proposed design exhibits the best global localization results. When compared to SP-SLAM, the proposed method leads to a 4.33% increase of trajectory estimation accuracy on 8 sub sequences of ICL-NUIM dataset and a 21.40% increase of trajectory estimation accuracy on 4 sub sequences of TUM dataset respectively.

关 键 词：室内RGB-D SLAM 特征基元 线特征配置 因子图 位姿估计 

分 类 号：U666.1[交通运输工程—船舶及航道工程]