基于场景显著区域的改进闭环检测算法

发布时间：2018-02-03 23:04

本文关键词： 同时定位与地图创建闭环检测逆向索引显著区域大视角变化　出处：《机器人》2017年01期 　论文类型：期刊论文

【摘要】：针对闭环路径不一致、闭环视角变化较大的VSLAM(视觉同时定位与地图创建)闭环检测问题,提出了一种基于场景显著区域的改进闭环检测算法.首先设计了基于场景显著区域的闭环检测架构,并构建了相应的闭环全概率模型.其次,提出了融合特征跟踪率与RGB直方图匹配的关键帧选取方法,降低了闭环检测过程的信息冗余.然后,采用基于逆向索引的预匹配场景选取方法,极大地提高了闭环检测的实时性.最后,根据本文的场景显著区域的几何特点给出了显著区域之间的观测几何匹配概率,提高了闭环检测算法对歧义场景的分辨能力.与FAB-MAP 2.0的对比实验表明,在保证较高的闭环准确率的条件下,本文算法能够有效提高对不重合路径闭环以及大视角变化的闭环的召回率,并具有良好的实时性.
[Abstract]:In the closed path inconsistent, closed loop VSLAM significant changes of view (visual simultaneous localization and mapping) closed-loop detection problem, and proposes an improved closed-loop scene detection algorithm based on salient region. The first design architecture scene Closed-loop Detection Based on salient regions, and construct the corresponding closed-loop probability model. Secondly, put forward key frame selection method of fusion feature tracking and RGB histogram matching, reduce the information redundancy of closed-loop detection process. Then, using the reverse index selection method based on the pre match scene, which greatly improves the real-time closed-loop detection. Finally, the matching probability between the observation geometry of salient regions are given according to the geometric characteristics of the scene significantly area, improve the resolution of the closed-loop detection method for ambiguous scenes. Compared with the FAB-MAP 2, higher in the quasi closed loop Under the condition of accuracy, this algorithm can effectively improve the closed loop recall of the closed loop and the large angle of view, and has good real-time performance.

【作者单位】：火箭军工程大学;
【分类号】：TP242;TP391.41
【正文快照】： 1引言(Introduction)闭环检测(loop closure detection)作为SLAM(同时定位与地图创建)问题的关键环节和基础问题,对消除机器人位姿估计的累积误差,降低地图不确定性至关重要[1,2].近年来,随着视觉SLAM技术的不断发展,基于视觉的闭环检测技术也有了长足的进步.由于闭环检测的应

【参考文献】