-
摘要:
同步定位与地图构建(SLAM)是视觉导航领域的关键技术之一,闭环检测是SLAM的基础问题。针对视觉SLAM闭环检测准确率不高的问题,提出一种高效准确的闭环检测算法。该算法由词袋模型、图像结构校验、跟踪预测模型3个模块构成。首先,将局部特征与全局特征相结合,设计了词袋模型与图像结构校验模块。词袋模型通过视觉单词比较图像之间的相似性,得到闭环候选帧。然后,图像结构校验模块灰度化、归一化当前图像与闭环候选图像。归一化之后的图像被直接作为局部特征的邻域,计算得到全局描述符,通过全局描述符判断闭环候选帧是否为有效的闭环。最后,针对传统闭环检测算法耗时随图像数量增加而显著增加的问题,设计了跟踪预测模块,以提高计算效率。实验中,与主流的DBoW算法相比,提出的闭环检测算法的准确率提升了20%以上,实时性也有更好的表现。
-
关键词:
- 同步定位与地图构建(SLAM) /
- 闭环检测 /
- 局部特征 /
- 全局特征 /
- 跟踪预测
Abstract:Simultaneous Localization and Mapping (SLAM) is one of the key technologies in visual navigation, and loop closure detection is a basis of SLAM. An efficient and accurate loop closure detection algorithm is proposed to solve the problem of low accuracy rate of SLAM loop closure detection. The loop closure detection algorithm consists of bag of words module, structure checking module, and tracking module. First, we design the bag of words model and structure checking module, combining local features with holistic features. The bag of words model compares the image similarities using visual words to obtain the closed-loop candidate frame. Then, structure checking module grayscales and normalizes the current image and the closed-loop candidate image. The normalized images are directly used as the patch of local feature to obtain holistic feature. Whether the closed-loop candidate frame is a valid closed loop is determined by the holistic descriptor. To address the problem that time consumption increases rapidly with the increase of image numbers, we design the tracking module to improve the computational efficiency. The comparative experiments with DBoW algorithm show that the proposed algorithm improves the accuracy by more than 20% and also has better real-time performance.
-
表 1 闭环检测准确率
Table 1. Loop closure detection accuracy
数据集标号 准确率/% 准确率提升/% DBoW算法 TPSV-DBoW算法 00 67.00 83.07 23.99 02 26.67 33.43 25.35 05 62.78 84.89 35.22 06 78.86 89.93 14.04 表 2 实时性对比
Table 2. Real-time performance comparison
数据集标号 图像数量 耗时/s 实时性提升/% DboW算法 TPSV-DboW算法 04 271 18.55 18.53 0.10 03 801 67.44 66.11 1.96 01、06、07 1 100 67.30 66.70 0.89 10 1 200 93.56 92.17 1.48 09 1 590 118.76 116.99 1.48 05 2 761 190.71 186.87 2.01 08 4 070 432.17 416.78 3.56 02 4 661 619.58 595.71 3.85 -
[1] GUI J J, GU D B, WANG S, et al.A review of visual inertial odometry from filtering and optimisation perspectives[J].Advanced Robotics, 2015, 29(20):1289-1301. doi: 10.1080/01691864.2015.1057616 [2] 何俊学, 李战明.基于视觉的同时定位与地图构建方法综述[J].计算机应用研究, 2010, 27(8):2839-2844. doi: 10.3969/j.issn.1001-3695.2010.08.007HE J X, LI Z M.Survey of vision-based approach to simultaneous localization and mapping[J].Application Research of Computers, 2010, 27(8):2839-2844(in Chinese). doi: 10.3969/j.issn.1001-3695.2010.08.007 [3] HESS W, KOHLER D, RAPP H H.Systems and methods of detecting loop closure in simultaneous localization and mapping (SLAM) applications: U.S.14/972, 938[P].2019-06-11. [4] CLEMENTE L, DAVISON A.Mapping large loops with a single hand-held camera[C]//Robotics Science and Systems, 2007, 2(2): 297-304. [5] WILLIAMS B, CUMMINS M, NEIRA J, et al.An image-to-map loop closing method for monocular SLAM[C]//IEEE International Conference on Intelligent Robots and Systems.Piscataway: IEEE Press, 2008: 2053-2059. [6] 刘强, 段富海.复杂环境下视觉SLAM闭环检测方法综述[J].机器人, 2018, 40(6):123-136.LIU Q, DUAN F H.A survey of loop-closure detection method of visual SLAM in complex environments[J].Robot, 2018, 40(6):123-136(in Chinese). [7] 刘国忠, 胡钊政.基于SURF和ORB全局特征的快速闭环检测[J].机器人, 2017, 39(1):36-45. doi: 10.3969/j.issn.1004-6437.2017.01.005LIU G Z, HU Z Z.Fast loop closure detection based on holistic features from SURF and ORB[J].Robot, 2017, 39(1):36-45(in Chinese). doi: 10.3969/j.issn.1004-6437.2017.01.005 [8] KROSE B J A, VLASSIS N, BUNSCHOTEN R, et al.A probabilistic model for appearance-based robot localization[J].Image & Vision Computing, 2001, 19(6):381-391. [9] LOWRY S M, WYETH G F, MILFORD M J.Unsupervised online learning of condition-invariant images for place recognition[J].Procedia-Social and Behavioral Sciences, 2014, 106:1418-1427. [10] CUMMINS M, NEWMAN P.Probabilistic appearance based navigation and loop closing[C]//Proceedings 2007 IEEE International Conference on Robotics and Automation.Piscataway: IEEE Press, 2007: 2042-2048. [11] GALVEZ-LOPEZ D, TARDOS J D.Bags of binary words for fast place recognition in image sequences[J].IEEE Transactions on Robotics, 2012, 28(5):1188-1197. doi: 10.1109/TRO.2012.2197158 [12] MUR-ARTAL R, TARDOS J D.ORB-SLAM2:An open-source SLAM system for monocular, stereo, and RGB-D cameras[J].IEEE Transactions on Robotics, 2017, 33(5):1255-1262. doi: 10.1109/TRO.2017.2705103 [13] 梁志伟, 陈燕燕, 朱松豪, 等.基于视觉词典的单目视觉闭环检测算法[J].模式识别与人工智能, 2013, 26(6):561-570. doi: 10.3969/j.issn.1003-6059.2013.06.007LIANG Z W, CHEN Y Y, ZHU S H, et al.Loop closure detection algorithm based on monocular vision using visual dictionary[J].Pattern Recognition and Artificial Intelligence, 2013, 26(6):561-570(in Chinese). doi: 10.3969/j.issn.1003-6059.2013.06.007 [14] GUCLU O, CAN A B.Fast and effective loop closure detection to improve SLAM performance[J].Journal of Intelligent and Robotic Systems, 2019, 93(3-4):495-517. doi: 10.1007/s10846-017-0718-z [15] ZHANG G, YAN X, YE Y.Loop closure detection via maximization of mutual information[J].IEEE Access, 2019, 7:124217-124232. doi: 10.1109/ACCESS.2019.2937967