北京航空航天大学学报 ›› 2017, Vol. 43 ›› Issue (4): 660-666.doi: 10.13700/j.bh.1001-5965.2016.0273

• 论文 • 上一篇    下一篇

单目SLAM直线匹配增强平面发现方法

蒙山, 唐文名   

  1. 深圳大学 信息工程学院, 深圳 518060
  • 收稿日期:2016-04-08 出版日期:2017-04-20 发布日期:2017-05-05
  • 通讯作者: 蒙山,E-mail:mengshan@szu.edu.cn E-mail:mengshan@szu.edu.cn
  • 作者简介:蒙山 男,博士,副教授,硕士生导师。主要研究方向:智能机器人系统运动建模与空间结构设计。;唐文名 男,硕士研究生。主要研究方向:三维环境建模、视觉导航。
  • 基金资助:
    国家“863”计划(2015AA042303);广东省科技计划(2015A030401016);深圳市基础研究项目(JCYJ20150629152510439)

Monocular SLAM plane discovery method enhanced by line segments matching

MENG Shan, TANG Wenming   

  1. College of Information Engineering, Shenzhen University, Shenzhen 518060, China
  • Received:2016-04-08 Online:2017-04-20 Published:2017-05-05

摘要: 针对微小型机器人及无人机系统日益迫切的轻量化视觉导航需求,提出了一种多维几何特征单目视觉三维环境建模方法。单一点特征单目SLAM制图方法地图描述效率相对较低,噪声容忍性能需要进一步提高。将线和面特征引入单目SLAM的三维地图构建过程,提高系统三维空间建模的搜索速度和稳定性。利用快速直线搜索算法,并基于二维直线匹配生成三维空间直线。现有基于三维空间特征点生成最小采样集的J-Linkage算法需要的倾向向量维数较高,完成单目SLAM常见场景三维平面聚类所需的计算量大。通过点线特征结合以及直线增强的J-Linkage算法可以提高特征平面聚类速度和稳定性,减少系统三维空间表达的冗余信息。

关键词: SLAM, 单目视觉, 几何聚类, 微小型机器人, 无人机

Abstract: To meet the self-navigation need of light-weighted robots, e.g. small UAV, we propose a multi-dimensional geometric feature extraction method for monocular SLAM. Feature points based SLAM mapping method is vulnerable to noisy samples and its description efficiency of complex environments needs to be increased. This method introduced the line and plane features to the three-dimensional map building process. It improved the monocular SLAM application system's key frame matching speed and overall stability. A rapid line matching algorithm was implemented, and three-dimensional lines were drawn by two-dimensional lines matching. Traditional space points based J-Linkage method drove its preference set's dimension high, and then remarkable calculation cost was needed for clustering points with multiple models, which is common during monocular SLAM mapping process. An enhanced J-Linkage algorithm was presented for feature plane extraction. With the combination of multi-dimensional geometric features, the reliability monocular SLAM system's mapping process was improved. The representative redundancy of the SLAM applications was reduced.

Key words: SLAM, monocular vision, geometrical clustering, light-weighted robots, UAV

中图分类号: 


版权所有 © 《北京航空航天大学学报》编辑部
通讯地址:北京市海淀区学院路37号 北京航空航天大学学报编辑部 邮编:100191 E-mail:jbuaa@buaa.edu.cn
本系统由北京玛格泰克科技发展有限公司设计开发