基于动态语义特征的视觉SLAM系统  

Visual SLAM System Based on Dynamic Semantic Features

作　　者：任伟建[1,2] 张志强 康朝海[1,2] 霍凤财[1,2] 孙勤江 陈建玲 REN Weijian;ZHANG Zhiqiang;KANG Chaohai;HUO Fengcai;SUN Qinjiang;CHEN Jianling(Department of Electrical and Information Engineering,Northeast Petroleum University,Daqing 163318,China;Heilongjiang Provincial Key Laboratory of Networking and Intelligent Control,Northeast Petroleum University,Daqing 163318,China;Tianjin Branch,China National Offshore Oil Corporation,Tianjin 300450,China)

机构地区：[1]东北石油大学电气信息工程学院,黑龙江大庆163318 [2]东北石油大学黑龙江省网络化与智能控制重点实验室,黑龙江大庆163318 [3]中海石油(中国)有限公司天津分公司,天津300450

出　　处：《吉林大学学报（信息科学版）》2023年第6期1041-1047,共7页Journal of Jilin University（Information Science Edition）

基　　金：国家自然科学基金资助项目(61933007,61873058)。

摘　　要：针对视觉SLAM(Simultaneous Localization and Mapping)在真实场景下出现动态物体(如行人,车辆、动物)等影响算法定位和建图精确性的问题,基于ORB-SLAM3(Oriented FAST and Rotated BRIEF-Simultaneous Localization and Mapping 3)提出了YOLOv3-ORB-SLAM3算法。该算法在ORB-SLAM3的基础上增加了语义线程,采用动态和静态场景特征提取双线程机制:语义线程使用YOLOv3对场景中动态物体进行语义识别目标检测,同时对提取的动态区域特征点进行离群点剔除;跟踪线程通过ORB特征提取场景区域特征,结合语义信息获得静态场景特征送入后端,从而消除动态场景对系统的干扰,提升视觉SLAM算法定位精度。利用TUM(Technical University of Munich)数据集验证,结果表明YOLOv3-ORB-SLAM3算法在单目模式下动态序列相比ORB-SLAM3算法ATE(Average Treatment Effect)指标下降30%左右,RGB-D(Red,Green and Blue-Depth)模式下动态序列ATE指标下降10%,静态序列未有明显下降。Aiming at the problems that dynamic objects(such as pedestrins,vehicles,animals) appear in visual SLAM(Simultaneous Localization and Mapping) in real scenes,affect the accuracy of algorithm positioning and mapping,the YOLOv3-ORB-SLAM3(Oriented FAST and Rotated BRIEF-Simultaneous Localization and Mapping 3) algorithm is proposed based on ORB-SLAM3.The algorithm adds a semantic thread on the basis of ORB-SLAM3,and the thread uses YOLOv3 to perform semantic recognition target detection on dynamic objects in the scene.The outliers are removed from the extracted feature points on the tracking thread,and the static environment area extracted by the ORB feature,thereby the positioning accuracy of the visual SLAM algorithm is improved.The TUM(Technical University of Munich)data set is used to verify the positioning accuracy of the algorithm in monocular and RGB-D(Red,Green and Blue-Depth) modes.The verification results show that the dynamic sequence of the YOLOv3-ORB-SLAM3 algorithm in monocular mode is about 30% lower than that of the ORB-SLAM3 algorithm in RGB-D mode,the dynamic sequence decreases by 10%,and the static sequence does not decrease significantly.

关 键 词：目标检测 ORB-SLAM3算法 动态场景 单目相机 深度相机 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]