摘要
针对三维重建系统使用ORB进行特征的提取并应用快速最近邻进行特征匹配，以及应用Shi-Tomasi角点检测的算法进行特征选择，从而实现位姿估计的数据关联，在此基础上应用ICP算法实现了系统的位姿初步计算。当匹配数对较少，系统基于点到面误差最小化的原则进行位姿计算，并在跟踪失败时进行重定位。这样做的优点是，系统不但可以应对特征丰富的场合，同时也可以处理特征数量较少的场景，提高了系统的鲁棒性。在位姿估计中还应用了单指令多数据流及多线程的支持，提高了运行的实时性能。位姿的初步计算完成后利用捆束优化的方法进行位姿的优化，并通过回环检测降低长时间运行的累积误差，提高了位姿估计的精度。
关键词： RGB-D；SLAM；特征提取；地图构建；位姿估计
Abstract
For the 3D reconstruction system, ORB is used for feature extraction, fast nearest neighbor is used for feature matching, and Shi-Tomasi corner detection algorithm is used for feature selection, so as to realize the data association of pose estimation. On this basis, the ICP algorithm is used to achieve Preliminary calculation of the pose of the system. When the matching logarithm is small, the system calculates the pose based on the principle of minimizing the point-to-surface error, and relocates when the tracking fails. The advantage of this is that the system can not only deal with situations with rich features, but also can handle scenes with a small number of features, which improves the robustness of the system. In the pose estimation, the support of single instruction, multiple data streams and multiple threads is also applied, which improves the real-time performance of the operation. After the preliminary calculation of the pose is completed, the bundle optimization method is used to optimize the pose, and the accumulated error of long-time running is reduced through loop detection, and the accuracy of the pose estimation is improved.
Key words： RGB-D; SLAM; Feature extraction; Map construction; Pose estimation
参考文献 References
[1] Liu H M, Zhang G F, Bao H J. A survey of monocular simultaneous localization and mapping[J].Journal of Computer-Aided Design & Computer Graphics, 2016, 6: 855-868.
[2] Smith R C, Cheeseman P. On the representation and estimation of spatial uncertainty[J]. The international journal of Robotics Research, 1986, 5: 56-68.
[3] Rublee E, Rabaud V, Konolige K, et al. ORB: An efficient alternative to SIFT or SURF[C]// 2011 International conference on computer vision. IEEE, 2011: 2564-2571.
[4] Viswanathan D G. Features from accelerated segment test (fast)[C]// Proceedings of the 10th workshop on Image Analysis for Multimedia Interactive Services, London, UK. 2009: 6-8.
[5] Harris C G, Stephens M. A combined corner and edge detector[C]// Alvey vision conference. 1988, 15: 10-5244.
[6] Shi J, Tomasi C. Good Features to Track[J]. Proceedings CVPR, IEEE Computer Society Conference on Computer Vision and Pattern Recognition, 2002, 600.
[7] Calonder M, Lepetit V, Strecha C, et al. Brief: Binary robust independent elementary features[C]// European conference on computer vision. Springer, Berlin, Heidelberg, 2010: 778-792.
[8] Pomerleau F, Colas F, Siegwart R . A Review of Point Cloud Registration Algorithms for Mobile Robotics[J]. Foundations & Trends in Robotics, 2015, 4: 1-104.
[9] Alonso I, Riazuelo L, Murillo A C. Enhancing v-slam keyframe selection with an efficient ConvNet for semantic analysis[C]// 2019 International Conference on Robotics and Automation (ICRA). IEEE, 2019: 4717-4723.
[10] 高翔, 张涛. 视觉SLAM十四讲：从理论到实践[M]. 北京: 电子工业出版社, 2019.
[11] Barron J T. A general and adaptive robust loss function[C]// Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2019: 4331-4339.
[12] Hertzberg C. A framework for sparse, non-linear least squares problems on manifolds[C]// Universität Bremen, 2008.
[13] Galvez-Lpez D , Tardos J D . Bags of Binary Words for Fast Place Recognition in Image Sequences[J]. IEEE Transactions on Robotics, 2012, 28: 1188-1197.