同步定位与建图技术的发展现状

Emerging development of simultaneous localization and mapping

同步定位与建图技术在增强现实、自动驾驶、智能机器人等领域具有重要作用。SLAM技术的目标是在给定传感器输入的情况下,实现对自身的定位,同时构建周围环境的地图。视觉SLAM技术是以摄像机等图像传感器为主要输入信息来源,相比于基于其他传感器的SLAM分支,具有成本低、拓展性强、信息丰富等优势,能够适应更广泛的场景、满足更多种类的需求,功能更多。本文首先讨论了视觉SLAM技术的背景与意义,阐述了SLAM算法的演变过程,分别论述了现阶段主流SLAM算法中不同模块的算法以及SLAM系统的硬件部署。同时,本文探讨了视觉SLAM与其他传感器融合组成的多传感器SLAM系统、多机协同的SLAM系统以及结合深度学习算法的SLAM系统的发展现状和不同算法的优缺点。最后,本文提出了目前SLAM系统研究与发展方向。

Simultaneous localization and mapping(SLAM)is one of critical techniques for applications including augmented reality(AR),automatic driving,and autonomous robots.SLAM aims to self-localize and to construct the map of surrounding environment leveraging the input of sensors such as camera,radar,and inertial measurement unit.As one major family of SLAM,visual SLAM(VSLAM)mainly relies on the image captured from the camera.Compared with SLAM based on other sensors,VSLAM has lower device cost,better extensibility,and richer information.VSLAM can be used widely in different environments and for various requirements.In this work,the development of VSLAM is discussed.Firstly,the background and history of VSLAM is analyzed.Then the development of the sub-modules of VSLAM are discussed respectively.Next,the hardware implementation techniques are discussed.Besides,the emerging VSLAM researches including multi-sensor SLAM,multi-agent SLAM,and SLAM with deep learning are discussed.Finally,we also provide some potential and promising researches.