[1]刘鸿勋,王 伟.双目相机和激光雷达的融合SLAM研究[J].南京师范大学学报(工程技术版),2021,(01):064-71.[doi:10.3969/j.issn.1672-1292.2021.01.010]
 Liu Hongxun,Wang Wei.Research on Fusion SLAM of Binocular Camera and Lidar[J].Journal of Nanjing Normal University(Engineering and Technology),2021,(01):064-71.[doi:10.3969/j.issn.1672-1292.2021.01.010]
点击复制

双目相机和激光雷达的融合SLAM研究
分享到:

南京师范大学学报(工程技术版)[ISSN:1006-6977/CN:61-1281/TN]

卷:
期数:
2021年01期
页码:
064-71
栏目:
计算机科学与技术
出版日期:
2021-03-15

文章信息/Info

Title:
Research on Fusion SLAM of Binocular Camera and Lidar
文章编号:
1672-1292(2021)01-0064-08
作者:
刘鸿勋王 伟
南京信息工程大学自动化学院,江苏 南京 210006
Author(s):
Liu HongxunWang Wei
College of Automation,Nanjing University of Information Science and Technology,Nanjing 210006,China
关键词:
FAST特征点图优化SLAM双目视觉词袋
Keywords:
FAST feature pointsgraph optimizedSLAMbinocular visionbag of words
分类号:
TP242.6
DOI:
10.3969/j.issn.1672-1292.2021.01.010
文献标志码:
A
摘要:
针对基于图优化的激光SLAM算法在高相似度的场景中闭环检测出错的问题,提出使用双目相机进行闭环检测的方法. 使用加入旋转不变性的FAST特征点和BRIEF描述子进行双目深度估计; 引入局部地图的概念,使用单帧激光雷达数据与局部地图进行匹配,提高SLAM前端的精度. 使用基于词袋(bag of words,BOW)模型的k叉树字典评估图片相似度从而完成闭环检测,最后构建全局优化问题并求解. 与主流开源激光雷达SLAM算法的对比实验表明,研究内容改善了只使用激光雷达数据进行闭环检测的方法在相似度较高场景下失效
Abstract:
Aiming at the problem of closed-loop detection errors in the graph-optimized laser SLAM algorithm in high-similarity scenes,the method of using binocular cameras for closed-loop detection is presented in this paper. We use FAST feature points and BRIEF descriptors with rotation invariance for binocular depth estimation. We introduce the concept of local maps,and use single-frame lidar data to match with local maps to improve the accuracy of the SLAM front-end. A k-ary tree dictionary based on the BOW model is used to evaluate the similarity of the pictures to complete the closed-loop detection,and finally,the global optimization problem is constructed and solved. The contrastive experiment with the mainstream open source lidar SLAM algorithms shows that the method improves the problem which the closed-loop detection method that only uses lidar data for closed-loop detection fails in a scene with high similarity,and that the operation effect in a larger area is significantly better than that based on Filtered SLAM algorithm.

参考文献/References:

[1] SEBASTIAN T. Probabilistic robotics[M]. Beijing:China Machine Press,2017.
[2]Aisha W B. Dynamic pose graph SLAM:long-term mapping in low dynamic environments[C]//Intelligent Robots and Systems. Vilamoura,Portugal,2012.
[3]张恒,徐万红,张禹. 激光雷达与深度相机融合的SLAM技术研究[J]. 机械工程师,2020,47(7):22-24.
[4]陈庆伟,李民东,罗川,等. 视觉SLAM中图像特征点提取与匹配算法研究[J]. 现代制造工程,2019,41(6):44-45.
[5]高翔,张涛. 视觉SLAM十四讲[M]. 北京:电子工业出版社,2017:132-138.
[6]刘国林,姜岩,陶华学. 非线性最小二乘参数平差[J]. 测绘学报,1998,27(3):224-230.
[7]吴玲,刘忠,卢发兴. 全局收敛高斯-牛顿法解非线性最小二乘定位问题[J]. 火控雷达技术,2003,32(1):3-4.
[8]KENSUKE K,KANJI T. Multi-scale bag-of-features for large-size map retrieval[C]//Robotics and Biomimeties. Tianjin:IEEE Computer Press,2010:961-966.
[9]SIVIC J,ZISSERMAN A. Video google:a text retrieval approach to object matching in videos[C]//Computer Vision:IEEE,2003:1470-1477.
[10]ROBERTSON S. Understanding inverse document frequency:on theoretical arguments for IDF[J]. Journal of Documentation,2004,60(5):503-520.
[11]NISTER D,STEWENIUS H. Scalable recognition with a vocabulary tree[C]//Computer Vision and Pattern Recognition:IEEE,2006:2161-2168.
[12]OLSON E B. Real-time correlative scan matching[C]//IEEE International Conference on Robotics and Automation:IEEE,2009.

备注/Memo

备注/Memo:
收稿日期:2020-08-08.
通讯作者:王伟,博士,教授,研究方向:无人机控制. E-mail:wangwei@aydrone.com
更新日期/Last Update: 2021-03-15