计算机应用研究2025，Vol.42Issue(6)：1909-1914,6.DOI:10.19734/j.issn.1001-3695.2024.07.0349
基于运动分割的动态SLAM联合优化算法
Dynamic SLAM joint optimization algorithm based on motion segmentation
摘要
Abstract
In order to address the challenge of traditional SLAM being easily disrupted by dynamic objects in the scene,this paper proposed a visual SLAM algorithm suitable for dynamic environments,referred to as GMS-SLAM.In this approach,it re-placed semantic segmentation with a geometric method based on epipolar constraints,and employed optical dilation to further eliminate potential coplanar ambiguities.It modeled dynamic objects using motion constraints and rigidity constraints.This app-roach utilized the graph optimization module to optimize the camera pose information,and transformed human skeletal mode-ling into rigid objects to be incorporated into the graph optimization process.Additionally,by exploiting the strong correlation between camera self-motion and epipolar constraints,it integrated the segmentation network and optimization module into a unified framework for joint refinement.Experimental validation on the KITTI and highly dynamic Shibuya datasets demon-strates that GMS-SLAM achieves substantial improvements in accuracy compared to state-of-the-art dynamic visual algorithms,such as DytanVO,and classical dynamic algorithms like DynaSLAM,exhibiting superior localization performance in dynamic environments.关键词
动态SLAM/图优化/运动分割/刚性约束Key words
dynamic SLAM/graph optimization/motion segmentation/rigid constrains分类
信息技术与安全科学引用本文复制引用
张宇曈,范馨月,周志远,谢源远..基于运动分割的动态SLAM联合优化算法[J].计算机应用研究,2025,42(6):1909-1914,6.基金项目
国家自然科学基金资助项目(62271096) （62271096）
