大规模环境下基于图优化SLAM的图构建方法

王忠立^1,2, 赵杰², 蔡鹤皋²

(1.北京交通大学 电子信息工程学院, 100044 北京; 2. 机器人技术与系统国家重点实验室(哈尔滨工业大学), 150080 哈尔滨)



摘要:

分析总结了基于图优化同步定位和地图构建(SLAM)前端图构建过程的各种方法. 对现有SLAM研究方法进行分类,指出基于Kalman滤波器、粒子滤波器、图优化方法的优缺点；重点介绍SLAM问题的3种图建模方法,即动态贝叶斯网络的图建模方法、基于因子图的建模方法、基于Markov随机场的建模方法；对图优化SLAM方法前端图构建的核心环节——帧间数据关联和环形闭合检测方法进行了分析；讨论了特征提取、特征匹配、运动估计、环形闭合检测等方面的最新研究成果.

关键词:  移动机器人  同步定位与建图  动态贝叶斯网络  图建模  数据关联

DOI：10.11918/j.issn.0367-6234.2015.01.012

分类号:TP242.6

基金项目:国家自然科学基金(61075079).



A survey of front end method for graph based slam under large scale environment

WANG Zhongli^1,2, ZHAO Jie², CAI Hegao²

（1．School of Electronic and Information Engineering, Beijing Jiaotong University, 100044 Beijing, China; 2．State Key Laboratory of Robotics and System, Harbin Institute of Technology, 150080 Harbin, China)

Abstract:

The existing graph-construction methods for graph optimization-based SLAM are summarized. The SLAM methods can be divided into three main classes, Kalman filter-based, partical filter-based and graph optimization-based, and the advantages and disadvantages of each class are overviewed. Moreover, there are mainly three graph modeling methods for the graph optimization-based SLAM problem, namely dynamic Bayesian network (DBN)-based model, factor graph-based model and Markov random field-based model. The key techniques of the front-end stage in graph optimization-based SLAM method, which mainly include data association between consecutive frame and loop closure detection, are discussed. Some newest research achievements on feature extraction, matching method, motion estimation, loop closure detection are introduced.

Key words:  mobile robot  simultaneous localization and mapping (SLAM)  dynamic Bayesian network  graph modeling  data association