重庆科技大学学报（自然科学版）2025，Vol.27Issue(5)：102-114,13.DOI:10.19406/j.issn.2097-4531.2025.05.012
动态环境下基于语义分割和LK光流法的视觉SLAM算法研究
Research on Visual SLAM Algorithm Based on Semantic Segmentation and LK Optical Flow in Dynamic Environments
摘要
Abstract
Visual SLAM in dynamic environments often suffers from reduced localization accuracy and ghosting arti-facts in dense point cloud maps.To address these issues,a dynamic visual SLAM algorithm that integrates semantic segmentation with the Lucas-Kanade(LK)optical flow method is proposed.Specifically,a dynamic feature point removal module based on YOLO11n-Seg is incorporated into the RTAB-Map framework to generate dynamic object masks and eliminate dynamic features;a filtering module based on LK optical flow method is introduced to further eliminate missed dynamic feature points;and dynamic object masks are utilized to remove ghosting artifacts in the dense point cloud map.The experimental results demonstrate that,compared to the RTAB-Map algorithm,the im-proved algorithm achieves average improvements of 77.62％and 72.59％in the ERMS and σ of the absolute trajecto-ry error on the high-dynamic sequences of the TUM RGB-D dataset;average improvements of 71.19％and 74.81％in the ERMS and σ of the relative pose error translational drift;and average improvements of 51.07％and 51.79％in the ERMS and σ of the rotational drift.Furthermore,the proposed method shows a clear advantage in eliminating ghosting artifacts in dense point cloud maps.关键词
同时定位与地图构建/动态特征点/语义分割/LK光流法/残影消除Key words
simultaneous localization and mapping/dynamic feature points/semantic segmentation/Lucas-Kanade optical flow/ghosting elimination分类
信息技术与安全科学引用本文复制引用
陈千禧,彭龑,何建华..动态环境下基于语义分割和LK光流法的视觉SLAM算法研究[J].重庆科技大学学报（自然科学版）,2025,27(5):102-114,13.基金项目
四川省智慧旅游研究基地资助项目(ZHYJ24-04) （ZHYJ24-04）
