面向动态环境的视觉惯性SLAM算法  

Visual inertial SLAM algorithm for dynamic environment

作　　者：崔林飞 黄丹丹 王祎旻 耿振野[1] 刘智[1] CUI Lin-fei;HUANG Dan-dan;WANG Yi-min;GENG Zhen-ye;LIU Zhi(School of Electronic Information Engineering,Changchun University of Science and Technology,Changchun 130000,China)

机构地区：[1]长春理工大学电子信息工程学院,吉林长春130000

出　　处：《计算机工程与设计》2022年第3期713-720,共8页Computer Engineering and Design

基　　金：吉林省自然科学基金项目(20200201167JC)。

摘　　要：针对传统的同时定位与建图(SLAM)算法在动态环境中会降低自身运动估计的精确性以及系统鲁棒性的问题,提出一种适用于动态环境的视觉惯性SLAM算法——DVI-SLAM(dynamic visual inertial SLAM)。根据对极几何约束检测并去掉动态特征,利用更加精确的静态特征进行状态估计;添加视觉信息自适应权重因子,提高系统的鲁棒性。改进的SLAM算法在公开的视觉惯性数据集TUM-VI上进行相关实验,实验结果与VINS-MONO相比在高动态场景中的定位精度平均提高了47.34%。To solve the problem that traditional simultaneous localization and mapping(SLAM)algorithm will reduce the accuracy of its ego-motion estimation and the robustness of the system in a dynamic environment,a visual-inertial SLAM algorithm suitable for dynamic environments(DVI-SLAM)was proposed.Dynamic features based on epipolar geometric constraint were detected and removed,more accurate static features were used for state estimation.Visual information adaptive weighting factors were added to improve the robustness of the system.Experiments were carried out for the proposed algorithm in the open visual-inertial data set TUM-VI.Compared with VINS-MONO,the experimental results show an average increase of 47.34%in positioning accuracy in high dynamic scenes.

关 键 词：同时定位与建图 动态环境 动态检测 多传感器融合 对极几何 

分 类 号：TP242[自动化与计算机技术—检测技术与自动化装置]