摘要
同步定位与建图(Simultaneous Localization and Mapping,SLAM)问题近年来已成为机器人导航领域的热门研究话题,作为其重要环节之一,回环检测用以消除整个过程中的累积误差.针对该环节高效率的需求,本文提出了一种基于局部特征ORB和全局描述符VLAD组合的快速回环检测算法.首先,使用一种全新的二值特征的VLAD量化算法(Binary-VLAD)提取全局特征,在保持描述符代表性的同时加快运行速度.然后,在全局粗搜索阶段,改进倒排索引结构,有效地减少了计算量和存储空间.其次,在几何验证阶段,使用一种基于空间相似性的偏移稳定模型,无需像RANSAC一样恢复基本矩阵,简捷高效.最后,在3个数据集上进行了验证实验,并与经典的词袋模型方法以及最新的基于深度学习的方法进行对比.实验结果表明,本文所提出的算法仅耗时19ms,明显优于经典的词袋模型算法,相比于最新的深度学习算法,时间效率更是提升近10倍,并且在保持100%准确率的前提下,召回率优于两者.
Simultaneous Localization and Mapping(SLAM)has become a hot research topic in the field of robot navigation in recent years.As one of the important links,loopclosure detection is used to eliminate the cumulative errors in the whole process.To meet the needs of high efficiency,we propose a fast loop closure detection algorithm based on the combination of local feature ORB and global vector VLAD.Firstly,a new Binary-VLAD algorithm is used to extract global features,which can maintain the representativeness of descriptors and speed up the operation.Then,in the global rough search stage,the inverted index structure is improved to reduce the computation and storage space effectively.Secondly,in the geometric verification stage,the migration stability model based on spatial similarity is adopted,which is simple and efficient without restoring the basic matrix like RANSAC.Finally,validation experiments are carried out on three data sets,and compared with the classical word bag model method and the latest deep learning-based method.Experimental results show that the proposed algorithm only consumes 19ms,which is significantly better than the classical word bag model algorithm.Compared with the latest deep learning algorithm,the time efficiency is improved by nearly 10 times,and the recall rate is better than both under the premise of maintaining 100%accuracy.
作者
赵珊
管启
丁德锐
魏国亮
尚朝辉
ZHAO Shan;GUAN Qi;DING De-rui;WEI Guo-liang;SHANG Chao-hui(School of Optical-Electrical and Computer Engineering,University of Shanghai for Science and Technology,Shanghai 200093,China;School of College of Science,University of Shanghai for Science and Technology,Shanghai 200093,China)
出处
《小型微型计算机系统》
CSCD
北大核心
2023年第6期1318-1323,共6页
Journal of Chinese Computer Systems
基金
国家自然科学基金项目(61973219,61933007)资助.
