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(simultaneous localization and mapping,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(robot operating system,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.
Indoor dynamic simultaneous localization and mapping based on object detection
In order to improve the accuracy and real-time of location and mapping in indoor dynamic scenes,a simultaneous localization and mapping(SLAM)system for indoor dynamic scenes based on target detection was proposed.Using the real-time property of target detection,the dynamic objects in the camera image are recognized on the basis of the traditional ORB-SLAM2 algorithm is combined with YOLOv5 target detection network,generating a dynamic recognition box.According to the dynamic feature point discrimination method,only the ORB feature points on the dynamic object in the recognition box are removed,and the camera position and orientation are estimated using the remaining feature points.Finally,the dense point cloud map and octree map containing only static objects are established.At the same time,the simulation is carried out under robot operating system(ROS),using Socket communication mode to replace the topic communication mode in ROS,combining ORB_SLAM2 algorithm with YOLOv5 target detection network,so as to improve the real-time performance of positioning and mapping.The results of many experiments on TUM data set show that compared with ORB-SLAM2 system,the system greatly improves the accuracy of camera pose and improves the processing speed of each frame tracking.
simultaneous localization and mapping systemtarget detectionindoor dynamic environmentORB feature pointpose estimationdense point cloud mapoctree maprobot operating system
国家科技期刊平台
NSTL
万方数据
维普
