软件导刊2025，Vol.24Issue(6)：142-150,9.DOI:10.11907/rjdk.241357
动态场景下基于图像分割与融合点线特征的视觉SLAM方法
Visual SLAM Method Based on Image Segmentation and Fusion of Point and Line Features in Dynamic Scenes
摘要
Abstract
Aiming at the problems of poor positioning accuracy and insufficient robustness of visual SLAM in dynamic scenes,a visual SLAM algorithm SPL-SLAM suitable for dynamic scenes is proposed.The algorithm first utilizes the YOLOv8 seg instance segmentation network to detect prior regions for dynamic region recognition and semantic map construction;Secondly,the robustness of the system is enhanced by in-troducing line feature constraints;Again,use motion consistency detection and angle detection algorithms to identify dynamic regions,and fil-ter dynamic point features and line features;Finally,an adaptive weight error function was designed and combined with residual static point and line features for pose estimation,while creating a static dense point cloud map to avoid dynamic object interference.Experimental verifica-tion was conducted using the TUM dataset and real scenes,and the results showed that compared to ORB-SLAM3 and other related dynamic SLAM algorithms,the SPL-SLAM algorithm has better positioning accuracy in dynamic scenes.关键词
视觉SLAM/图像分割/位姿估计/点云地图/动态场景Key words
visual SLAM/image segmentation/pose estimation/point cloud map/dynamic scene分类
信息技术与安全科学引用本文复制引用
姚东昊,袁野,吴宝磊,刘娜..动态场景下基于图像分割与融合点线特征的视觉SLAM方法[J].软件导刊,2025,24(6):142-150,9.基金项目
国家重点研发计划项目(2023YFC3605800) （2023YFC3605800）
