摘要
针对于动态环境中同时定位与建图(SLAM)位姿估计精度低、鲁棒性差及实时性弱的问题,提出一种动态视觉惯性SLAM算法(DVI-SLAM):通过引入惯性测量单元(IMU)的运动先验信息对图像进行校正,显著提升相机大幅运动时的目标检测精度;结合语义信息与运动检测一致性,在快速获取静态特征点的同时,可保持较高的精度;为应对高度动态场景中基于语义的动态SLAM的挑战,利用来自IMU的运动信息估计动态物体运动状态,并最大限度地利用环境中的静态特征点。仿真结果表明,DVI-SLAM在室内高动态场景中的绝对轨迹误差(ATE)均方根值(RMSE)相比ORBSLAM3可降低96.2%,在室外动态环境中可降低59.9%;与现有主流算法相比,提出的方法能在动态室内外场景中表现出较高的定位精度和鲁棒性。
Aiming at the problem t of low pose estimation accuracy,poor robustness and weak real-time performance in simultaneous localization and mapping(SLAM)under dynamic environments,the paper proposed a dynamic visual-inertial SLAM algorithm(DVI-SLAM):the motion prior information was incorporated from the inertial measurement unit(IMU)to perform image correction,for improving target detection accuracy during rapid camera movements;through the integration of semantic information with motion consistency detection,the rapid identification of static feature points was realized without compromising localization precision;furthermore,to tackle the challenges of semantic-based dynamic SLAM in highly dynamic scenarios,the motion information was used from the IMU to estimate the motion states of dynamic objects,and maximize the utilization of static features in the environment.Simulation results showed that the proposed method could reduce the absolute trajectory error(ATE)root mean square error(RMSE)by 96.2%compared to ORB-SLAM3 in indoor highly dynamic scenarios,and by 59.9%in outdoor dynamic environments,indicating that,compared to existing mainstream algorithms,the proposed method would exhibit good improvements in localization accuracy and superior robustness in dynamic indoor and outdoor environments.
作者
范金龙
宁一鹏
王坚
郭郑伟
高菲菲
王薇薇
柴大帅
FAN Jinlong;NING Yipeng;WANG Jian;GUO Zhengwei;GAO Feifei;WANG Weiwei;CHAI Dashuai(School of Surveying and Geo-Informatics,Shandong Jianzhu University,Jinan 250101,China;School of Geomatics and Urban Spatial Informatics,Beijing University of Civil Engineering and Architecture,Beijing 102616,China;Qilu Institute of Aerospace Information,Jinan 250101,China)
出处
《导航定位学报》
北大核心
2025年第4期136-145,共10页
Journal of Navigation and Positioning
基金
国家自然科学基金项目(42204011)
山东省自然科学基金项目(ZR2021QD058,ZR2022QD108)。
