基于描述符辅助光流跟踪匹配的数据关联方法

夏华佳; 章红平; 陈德忠; 李团

doi:10.3963/j.jssn.1674-4861.2021.06.018

基于描述符辅助光流跟踪匹配的数据关联方法

doi: 10.3963/j.jssn.1674-4861.2021.06.018

武汉大学卫星导航定位技术研究中心武汉 430079

基金项目:

国家重点研发计划项目 SQ2018YFE020091

长江勘测规划设计研究院开放创新基金项目 CX2020K04

详细信息

作者简介:
夏华佳(1996—), 硕士研究生. 研究方向: 视觉/INS组合导航. E-mail: 67010973@qq.com

通讯作者:
章红平(1977—), 博士, 教授. 研究方向: GNSS/INS组合导航. E-mail: hpzhang@whu.edu.cn

中图分类号: V249.32+8
计量
- 文章访问数: 903
- HTML全文浏览量: 411
- PDF下载量: 18
- 被引次数: 0
出版历程
- 收稿日期: 2021-07-06
- 网络出版日期: 2022-01-12

A Data Association Method Based on Tracking and Matching of Assisted Optical Flows from the Descriptor

GNSS Research Center, Wuhan University, Wuhan 430079, China

摘要

摘要: 针对采用多状态约束卡尔曼滤波(MSCKF)的视觉惯性里程计定位精度易受特征点匹配异常值影响问题, 提出了1种基于描述符辅助光流跟踪匹配的数据关联方法。该方法采用金字塔LK光流对序列图像中特征点进行跟踪匹配, 计算每一对匹配点的rBRIEF描述符, 根据Hamming距离对描述符的相似度进行判断消除异常匹配点。在实验中从特征点匹配主观效果以及定位精度2个方面评估本文方法的有效性, 结果表明: 所提出方法能够有效滤除动态场景下图像特征匹配的异常值, 使用该方法处理后的图像进行MSCKF运动解算, 位置结果漂移率小于0.38%, 相较于未剔除异常匹配值的MSCKF算法结果, 改善了54.7%, 单帧图像处理时间约为39 ms。
- 视觉惯性 /
- 数据关联方法 /
- 特征点匹配 /
- rBRIEF描述符 /
- 光流跟踪 /
- MSCKF
Abstract: The positioning accuracy of visual inertial odometer using multi-state constrained Kalman filter(MSCKF)is easily affected by mismatching points. A data association method is proposed for mitigating these outliers in this study. First, pyramid Lucas-Kanade(LK)optical flow is used to track and match the features among the sequence images. Second, the rBRIEF descriptors of each pair of matching points are achieved. Third, the Hamming distances between two rBRIEF descriptors can be calculated. Furthermore, the similarity of these descriptors is then evaluated according to Hamming distance. Last, the matching points of low similarity are eliminated as outliers in the data processing. The performances of the proposed method is assessed by the effectiveness of matching and positioning accuracy of the feature point. The results indicate that the proposed method can eliminate mismatching points in dynamic image processing. The outlier-eliminated images are applied for the MSCKF motion estimation. The derived drift rate of positioning result is less than 0.38% and shows an improvement of 54.7% with no outlier-eliminated MSCKF algorithm. The single-frame image processing time is about 39 ms.
- visual-inertial /
- data association method /
- feature matching /
- RBRIEF descriptor /
- optical flow tracking /
- MSCKF

HTML全文

图 1 基于特征点的数据关联方案

Figure 1. Data association scheme based on feature points

下载: 全尺寸图片幻灯片

图 2 FAST角点检测模板

Figure 2. FAST corner detection template

下载: 全尺寸图片幻灯片

图 3 rBRIEF描述符旋转不变性

Figure 3. Rotation invariance of rBRIEF descriptor

下载: 全尺寸图片幻灯片

图 4 MSCKF视觉惯性里程计算法框图

Figure 4. MSCKF-based visual-inertial odometer algorithm

下载: 全尺寸图片幻灯片

图 5 特征点跟踪匹配点对数量图

Figure 5. Quantity of feature points tracking matching-point pairs

下载: 全尺寸图片幻灯片

图 6 载体导航状态

Figure 6. Navigation status of the carrier

下载: 全尺寸图片幻灯片

图 7 rBRIEF描述符滤除异常匹配前后对比

Figure 7. Comparison before and after rBRIEF descriptor filtering exception matching

下载: 全尺寸图片幻灯片

图 8 真值与不同算法解算轨迹图

Figure 8. Solving trajectory of the true value and different algorithms

下载: 全尺寸图片幻灯片

图 9 2种算法解算定位结果误差曲线图

Figure 9. Two algorithms solving the error curves of positioning results

下载: 全尺寸图片幻灯片

表 1 惯导设备参数

Table 1. Equipment parameters of inertial-navigation

指标	M40	POS320
陀螺零偏稳定性(℃/h)	50	0.3
角度随机游走(℃/$ \sqrt h $)	0.4	0.03
加表零偏稳定性(mGal)	200	200
速度随机游走/(m/(s/$ \sqrt h $))	0.1	0.05

下载: 导出CSV

表 2 Basler acA1600-20gm相机参数

Table 2. Parameters of Basler acA1600-20gm camera

传感器类型	CCD
传感器尺寸/mm	7.16×5.44
分辨率/pixie	1628×1236
像元大小/μm	4.4×4.4
接口	千兆网
曝光控制	相机API编程控制外部脉冲信号触发
帧率/(帧/s)	20
通道数/(颜色)	单通道/(黑白)
像素位深度/bits	12

下载: 导出CSV

表 3 VIO1.0解算误差统计表

Table 3. Statistics of VIO1.0 solution errors

	N/m	E/m	D/m	ROLL/(°)	PITCH/(°)	YAW/(°)
MAX	7.74	16.80	2.63	0.18	0.17	1.46
RMS	3.13	7.48	1.02	0.05	0.08	0.82

下载: 导出CSV

表 4 VIO2.0解算误差统计表

Table 4. Statistics of VIO2.0 solution errors

	N/m	E/m	D/m	ROLL/(°)	PITCH/(°)	YAW/(°)
MAX	5.51	7.17	2.43	0.14	0.13	1.05
RMS	2.07	3.26	1.03	0.03	0.05	0.45

下载: 导出CSV

参考文献(17)

[1]	姚卓, 章红平. 里程计辅助车载GNSS/INS组合导航性能分析[J]. 大地测量与地球动力学, 2018, 38(2): 206-210. https://www.cnki.com.cn/Article/CJFDTOTAL-DKXB201802024.htm YAO Zhuo, ZHANG Hongping. Performance analysis of odometer assisted vehicle GNSS/INS integrated navigation[J]. Journal of Geodesy and Geodynamics, 2018, 38(2): 206-210. (in Chinese). https://www.cnki.com.cn/Article/CJFDTOTAL-DKXB201802024.htm
[2]	QIU Xiaochen, ZHANG Hai, FU Wenxing, et al. Monocular visual-inertial odometry with an unbiased linear system model and robust feature tracking front-end[J]. Sensors, 2019, 19(8): 1941. doi: 10.3390/s19081941
[3]	DELMERICO J, SCARAMUZZA D. A benchmark comparison of monocular visual-inertial odometry algorithms for flying robots[C]. 2018 International IEEE Conference on Robotics and Automation, Queensland, Australia: IEEE, 2018.
[4]	FAESSLER M, FONTANA F, FORSTER C, et al. Autonomous, vision-based flight and live dense 3D mapping with a quadrotor micro aerial vehicle[J]. Journal of Field Robotics, 2016, 33(4): 431-450. doi: 10.1002/rob.21581
[5]	BLOESCH M, BURRI M, OMARI S, et al. Iterated extended Kalman filter based visual-inertial odometry using direct photometric feedback[J]. The International Journal of Robotics Research, 2017, 36(10): 1053-1072. doi: 10.1177/0278364917728574
[6]	MOURIKIS A I, ROUMELIOTIS S I. A multi-state constraint Kalman filter for vision-aided inertial navigation[C]. 2007 International IEEE Conference on Robotics and Automation, Roma, Italy: IEEE, 2007.
[7]	齐乃新, 杨小冈, 李小峰, 等. 基于ORB特征和LK光流的视觉里程计算法[J]. 仪器仪表学报, 2018, 39(12): 216-227. https://www.cnki.com.cn/Article/CJFDTOTAL-YQXB201812025.htm QI Naixin, YANG Xiaogang, LI Xiaofeng, et al. Visual odometry algorithm based on ORB features and LK optical flow[J]. Chinese Journal of Scientific Instrument, 2018, 39(12): 216-227. (in Chinese). https://www.cnki.com.cn/Article/CJFDTOTAL-YQXB201812025.htm
[8]	LIU Jingneng, ZENG Guihua. Improved global context descriptor for describing interest regions[J]. Journal of Shanghai Jiaotong University(Science), 2012, 17(2): 147-152. doi: 10.1007/s12204-012-1244-6
[9]	成怡, 朱伟康, 徐国伟. 基于余弦相似度的改进ORB匹配算法[J]. 天津工业大学学报, 2021, 40(1): 60-66. https://www.cnki.com.cn/Article/CJFDTOTAL-TJFZ202101011.htm CHENG Yi, ZHU Weikang, XU Guowei. An improved ORB matching algorithm based on cosine similarity[J]. Journal of Tianjin Polytechnic University, 2021, 40(1): 60-66. (in Chinese). https://www.cnki.com.cn/Article/CJFDTOTAL-TJFZ202101011.htm
[10]	邹斌, 赵小虎, 尹智帅. 基于改进ORB的图像特征匹配算法研究[J]. 激光与光电子学进展, 2021, 58(2): 96-103. https://www.cnki.com.cn/Article/CJFDTOTAL-JGDJ202102009.htm ZOU Bin, ZHAO Xiaohu, YIN Zhishuai. Image feature matching algorithm based on improved ORB[J]. Laser & Optoelectronics Progress, 2021, 58(2): 96-103. (in Chinese). https://www.cnki.com.cn/Article/CJFDTOTAL-JGDJ202102009.htm
[11]	王亭亭, 蔡志浩, 王英勋. 无人机室内视觉/惯导组合导航方法[J]. 北京航空航天大学学报, 2018, 44(1): 176-186. https://www.cnki.com.cn/Article/CJFDTOTAL-BJHK201801021.htm WANG Tingting, CAI Zhihao, WANG Yingxun. UAV indoor vision/inertial navigation integrated navigation method[J]. Journal of Beijing University of Aeronautics and Astronautics, 2018, 44(1): 176-186. (in Chinese). https://www.cnki.com.cn/Article/CJFDTOTAL-BJHK201801021.htm
[12]	刘奕杉. 一种基于改进光流法的视觉惯性状态估计器[J]. 科学技术创新, 2021(10): 17-18. https://www.cnki.com.cn/Article/CJFDTOTAL-HLKX202110008.htm LIU Yishan. A visual inertial state estimator based on improved optical flow method[J]. Scientific and Technological Innovation, 2021(10): 17-18. (in Chinese). https://www.cnki.com.cn/Article/CJFDTOTAL-HLKX202110008.htm
[13]	BOUGUET J Y. Pyramidal implementation of the affine lucas kanade feature tracker description of the algorithm[J]. Intel Corporation, 2001, 4(5): 1-10.
[14]	RUBLEE E, RABAUD V, KONOLIGE K, et al. ORB: An efficient alternative to SIFT or SURF[C]. 2011 International Conference on Computer Vision, Barcelona, Spain: IEEE, 2011.
[15]	CHUM O, MATAS J, KITTLER J. Locally optimized RANSAC[C]. Joint Pattern Recognition Symposium, Magdeburg, Germany: DAGM, 2003.
[16]	李团. 单频多模GNSS/INS/视觉紧组合高精度位姿估计方法研究[D]. 武汉: 武汉大学, 2019. LI Tuan. Research on the tightly coupled single-frequency multi-GNSS/INS/Vision integration for precise position and orientation estimation[D]. Wuhan: Wuhan University, 2019. (in Chinese).
[17]	MUR-ARTAL R, TARDÓS J D. Orb-slam2: An open-source slam system for monocular, stereo, andrgb-d cameras[J]. IEEE Transactions on Robotics, 2017, 33(5): 1255-1262. http://ieeexplore.ieee.org/document/7946260