基于单目视觉的同时定位与建图算法研究综述*

与传统基于激光传感器的同时定位与建图（SLAM）方法相比，基于图像视觉传感器SLAM方法能廉价的获得更多环境信息，帮助移动机器人提高智能性。不同于用带深度信息的3D传感器研究SLAM问题，单目视觉SLAM算法用二维图像序列在线构建三维环境地图并实现实时定位。针对多种单目视觉SLAM算法进行对比研究，分析了近10年来流行的单目视觉定位算法的主要思路及其分类，指出基于优化方法正取代滤波器方法成为主流方法。从初始化、位姿估计、地图创建、闭环检测等功能组件的角度分别总结了当下流行的各种单目视觉 SLAM 或Odometry系统的工作原理和关键技术，阐述它们的工作过程和性能特点。总结了近年最新单目视觉定位算法的设计思路，最后概括指出本领域的研究热点与发展趋势。

A Survey on Monocular Visual SLAM Algorithms

Compare with traditional laser scanner based SLAM method, camera based SLAM algorithms outperform in cost as well as information, and can make mobile robot smarter. Instead of using 3D sensors, monocular visual SLAM utilize 2D image sequence to reconstruct 3D map and perform real-time localization. This article aims at providing a survey on monocular visual SLAM algorithms. In this paper, we studied the most popular V-SLAM algorithms in the last decade and discussed the main principal and their classification; then point out that the optimization method will be the mainstream. It summarized the principal and practice of the state of the art visual SLAM or visual Odometry systems from the angle of initialization, pose estimation, map generation and loop closure. According to above, it also elaborates the design scheme and performance of the state of art monocular SLAM algorithms. At last, the article concludes with our viewpoint for the research trend.