YGL-SLAM:动态场景下基于点和线的语义SLAM系统  

YGL-SLAM:Point and Line Based Semantic SLAM System for Dynamic Scenes

作　　者：戴康佳 徐慧英[1] 朱信忠[1,2] 李悉钰 黄晓 陈国强[4] 张志雄 DAI Kangjia;XU Huiying;ZHU Xinzhong;LI Xiyu;HUANG Xiao;CHEN Guoqiang;ZHANG Zhixiong(College of Computer Science and Technology,Zhejiang Normal University,Jinhua 321004,Zhejiang,China;Beijing Geekplus Technology Co.,Ltd.,Beijing 100012,China;College of Education,Zhejiang Normal University,Jinhua 321004,Zhejiang,China;Zhejiang Rainbow Aerospace Measurement&Control Technology Co.,Ltd.,Hangzhou 311200,Zhejiang,China;College of Computer Science and Technology,Zhejiang Gongshang University,Hangzhou 310018,Zhejiang,China)

机构地区：[1]浙江师范大学计算机科学与技术学院,浙江金华321004 [2]北京极智嘉科技股份有限公司,北京100012 [3]浙江师范大学教育学院,浙江金华321004 [4]浙江航天润博测控技术有限公司,浙江杭州311200 [5]浙江工商大学计算机科学与技术学院,浙江杭州310018

出　　处：《计算机工程》2025年第3期95-104,共10页Computer Engineering

基　　金：国家自然科学基金(62376252);浙江省自然科学基金重点项目(LZ22F030003);国家级大学生创新训练计划重点项目(202310345042)。

摘　　要：传统的视觉同步定位与建图(SLAM)系统是基于静态环境这一假设的,然而在现实场景中往往存在动态物体,这可能导致SLAM位姿估计和地图构建的精度下降、鲁棒性变差,甚至出现跟踪丢失的情况。针对上述问题,基于ORB-SLAM2提出新的语义SLAM系统(YGL-SLAM)。该系统首先使用轻量级目标检测算法YOLOv8n追踪动态对象,获得动态对象的语义信息。然后在跟踪线程的同时提取点特征和线特征,根据获取的语义信息利用Z-score和对极几何算法剔除动态特征,以改进SLAM在动态场景中的表现。此外,鉴于轻量级目标检测算法在追踪动态对象时存在连续帧的漏检测问题,设计了基于相邻帧的检测补偿方法。在公开数据集TUM和Bonn上的测试结果表明,相比ORB-SLAM2,YGL-SLAM系统准确率提升超过90%,对比其他动态SLAM,YGL-SLAM也具有较高的准确度和鲁棒性。Traditional vision Simultaneous Localization And Mapping(SLAM)systems are based on the assumption of a static environment.However,real scenes often have dynamic objects,which may lead to decreased accuracy,deterioration of robustness,and even tracking loss in SLAM position estimation and map construction.To address these issues,this study proposes a new semantic SLAM system,named YGL-SLAM,based on ORB-SLAM2.The system first uses a lightweight target detection algorithm named YOLOv8n,to track dynamic objects and obtain their semantic information.Subsequently,both point and line features are extracted from the tracking thread,and the dynamic features are culled based on the acquired semantic information using the Z-score and parapolar geometry algorithms to improve the performance of SLAM in dynamic scenes.Given that lightweight target detection algorithms suffer from missed detection in consecutive frames when tracking dynamic objects,this study designs a detection compensation method based on neighboring frames.Testing on the public datasets TUM and Bonn reveals that YGL-SLAM system improves detection performance by over 90%compared to ORB-SLAM2,while demonstrating superior accuracy and robustness compared to other dynamic SLAM.

关 键 词：动态场景 语义同步定位与建图 线特征 深度学习 YGL-SLAM系统 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]