DOI:10.11991/yykj.202304030网络出版地址：https://link.cnki.net/urlid/23.1191.U.20231218.0917.002基于目标检测的室内动态场景同步定位与建图郭培涛，席志红哈尔滨工程大学信息与通信工程学院，黑龙江哈尔滨150001摘要：为了提高室内动态场景下定位与建图的准确性与实时性，提出了一种基于目标检测的室内动态场景同步定位与建图（simultaneouslocalizationandmapping,SLAM）系统。利用目标检测的实时性，在传统ORB_SLAM2算法上结合YOLOv5目标检测网络识别相机图像中的动态物体，生成动态识别框，根据动态特征点判别方法只将识别框内动态物体上的ORB特征点去除，利用剩余特征点进行相机位姿的估计，最后建立只含静态物体的稠密点云地图与八叉树地图。同时在机器人操作系统（robotoperatingsystem,ROS）下进行仿真，采用套接字（Socket）通信方式代替ROS中话题通信方式，将ORB_SLAM2算法与YOLOv5目标检测网络相结合，以提高定位与建图的实时性。在TUM数据集上进行多次实验结果表明，与ORB_SLAM2系统相比，本文系统相机位姿精确度大幅度提高，并且提高了每帧跟踪的处理速度。关键词：定位与建图系统；目标检测；室内动态环境；ORB特征点；位姿估计；稠密点云地图；八叉树地图；机器人操作系统中图分类号：TP242文献标志码：A文章编号：1009−671X(2024)02−0076−07IndoordynamicsimultaneouslocalizationandmappingbasedonobjectdetectionGUOPeitao,XIZhihongCollegeofInformationandCommunicationEngineering,HarbinEngineeringUniversity,Harbin150001,ChinaAbstract:Inordertoimprovetheaccuracyandreal-timeoflocationandmappinginindoordynamicscenes,asimultaneouslocalizationandmapping(SLAM)systemforindoordynamicscenesbasedontargetdetectionwasproposed.Usingthereal-timepropertyoftargetdetection,thedynamicobjectsinthecameraimagearerecognizedonthebasisofthetraditionalORB-SLAM2algorithmiscombinedwithYOLOv5targetdetectionnetwork,generatingadynamicrecognitionbox.Accordingtothedynamicfeaturepointdiscriminationmethod,onlytheORBfeaturepointsonthedynamicobjectintherecognitionboxareremoved,andthecamerapositionandorientationareestimatedusingtheremainingfeaturepoints.Finally,thedensepointcloudmapandoctreemapcontainingonlystaticobjectsareestablished.Atthesametime,thesimulationiscarriedoutunder...
