摘要
提出于2015年的ORB-SLAM是当今同时定位与地图创建(SLAM)系统中做得非常完善、非常易用的系统之一,代表着主流的特征点SLAM的一个高峰。系统包含了SLAM所有的主要模块:特征匹配、跟踪、建图、重定位和闭环检测,同时有单目、双目、RGBD相机、鱼眼相机的接口,能够实时计算出相机的轨迹,并生成场景的稀疏三维重建结果。ORB-SLAM3于2020年发布,集成了共视图、词袋、ORB特征三个手段进一步使定位更加准确。本文侧重结合最新的ORB-SLAM3版本,从相机模型、多地图系统以及地图融合三个主要方面对其最新发展进行介绍。对每一个方面,阐述其每一个版本的关键设计及改进,为ORB-SLAM系统提供一个全面的概述。对相关系统的研究者来说,掌握了解其发展是非常友好的。最后,本文介绍ORB-SLAM最新研究进展并做出评价。
Introduced in 2015, ORB-SLAM is one of the most well-developed and easy-to-use systems among current simultaneous localization and mapping (SLAM) systems, it reaches the peak of popular feature-based SLAM systems. This system contains all necessary modules of SLAM: feature matching, tracking, mapping, relocalization and loop closing. It is suitable for monocular camera, stereo camera, RGBD camera and fisheye camera. Also, it solves the trajectory of the camera in real time and generates sparse 3D reconstruction of scenes. ORB-SLAM3 was released in 2020, integrating three methods of covisibility graph, words bag and ORB feature matching. In this paper, three main aspects: camera model, multi-map system and map merging are discussed. For each aspect, its key techniques in each release are analyzed and improvements are demonstrated to provide a comprehensive overview of the ORB-SLAM system. It is very friendly for the researcher of relevant field to understand its developments. In the end, the latest research progress of ORB-SLAM is introduced and evaluated.
出处
《软件工程与应用》
2023年第1期78-86,共9页
Software Engineering and Applications
