• 中国计算机学会会刊
  • 中国科技核心期刊
  • 中文核心期刊

计算机工程与科学

• 图形与图像 • 上一篇    下一篇

线特征融合光流的单目SLAM算法

贾哲,冷建伟   

  1. (天津理工大学电气电子工程学院,天津 300384)
  • 收稿日期:2018-01-12 修回日期:2018-05-29 出版日期:2018-12-25 发布日期:2018-12-25

A monocular-SLAM algorithm based on
fusion of line feature and optical flow
 

JIA Zhe,LENG Jianwei   

  1. (School of Electrical and Electronic Engineering,Tianjin University of Technology,Tianjin 300384,China)
  • Received:2018-01-12 Revised:2018-05-29 Online:2018-12-25 Published:2018-12-25

摘要:

针对移动机器人的定位与建图问题,提出了基于图优化的单目线特征融合光流的同时定位和地图构建(SLAM)的算法。首先,针对主流视觉SLAM算法因采用点作为特征而导致构建的点云地图稀疏、难以准确表达环境结构信息等缺点,采用直线作为特征来构建地图,并采用图优化方法来提高定位精度和地图构建的准确性。然后,针对定位系统的处理速度很难达到实时性要求,将光流法引入以达到实时定位的效果。实验表明,基于线特征的地图构建有较高的建图精度,并且融合算法克服了光流法定位精度差和特征法处理速度慢的缺点,可提供较准确的实时定位输出,并对光照变化和场景纹理较少的情况有一定的鲁棒性。
 

关键词: 移动机器人, 单目SLAM, 线特征, 图优化, 光流

Abstract:

In order to solve the problem of localization and mapping of mobile robots, we propose a new simultaneou localization and mapping (SLAM) algorithm based on the fusion of line feature and optical flow. Firstly, the mainstream visual SLAM algorithms use points as features, resulting in that the point cloud map is sparse and it is difficult to accurately express the environmental structure information. Aiming at this problem, we take straight lines instead of points as a feature to construct the map, and employ the graph optimization method to improve the accuracy of localization and map construction. Secondly, the processing speed of the localization system cannot achieve realtime requirement, we introduce the optical flow method to realize realtime localization. Experimental results demonstrate that the map construction based on line features has higher mapping accuracy, and the fusion algorithm overcomes the shortcomings of poor localization accuracy of the optical flow and the low processing speed of the feature matching method, providing more accurate realtime localization output. And it is robust to circumstances such as illumination change and low scene texture.
 

Key words: mobile robot, monocular SLAM, line feature, graph optimization, optical flow