2.845

2023影响因子

(CJCR)

  • 中文核心
  • EI
  • 中国科技核心
  • Scopus
  • CSCD
  • 英国科学文摘

留言板

尊敬的读者、作者、审稿人, 关于本刊的投稿、审稿、编辑和出版的任何问题, 您可以本页添加留言。我们将尽快给您答复。谢谢您的支持!

姓名
邮箱
手机号码
标题
留言内容
验证码

基于深度匹配的由稀疏到稠密大位移运动光流估计

陈震 张道文 张聪炫 汪洋

陈震, 张道文, 张聪炫, 汪洋. 基于深度匹配的由稀疏到稠密大位移运动光流估计. 自动化学报, 2022, 48(9): 2316−2326 doi: 10.16383/j.aas.c190716
引用本文: 陈震, 张道文, 张聪炫, 汪洋. 基于深度匹配的由稀疏到稠密大位移运动光流估计. 自动化学报, 2022, 48(9): 2316−2326 doi: 10.16383/j.aas.c190716
Chen Zhen, Zhang Dao-Wen, Zhang Cong-Xuan, Wang Yang. Sparse-to-dense large displacement motion optical flow estimation based on deep matching. Acta Automatica Sinica, 2022, 48(9): 2316−2326 doi: 10.16383/j.aas.c190716
Citation: Chen Zhen, Zhang Dao-Wen, Zhang Cong-Xuan, Wang Yang. Sparse-to-dense large displacement motion optical flow estimation based on deep matching. Acta Automatica Sinica, 2022, 48(9): 2316−2326 doi: 10.16383/j.aas.c190716

基于深度匹配的由稀疏到稠密大位移运动光流估计

doi: 10.16383/j.aas.c190716
基金项目: 国家自然科学基金(61866026, 61772255), 江西省优势科技创新团队(20165BCB19007), 江西省杰出青年人才计划(20192BCB23011), 航空科学基金(2018ZC56008)资助
详细信息
    作者简介:

    陈震:南昌航空大学测试与光电工程学院教授. 主要研究方向为图像检测与智能识别, 计算机视觉.E-mail: dr_chenzhen@163.com

    张道文:南昌航空大学测试与光电工程学院硕士研究生. 主要研究方向为图像处理与模式识别.E-mail: daowenzhang@163.com

    张聪炫:南昌航空大学测试与光电工程学院教授. 主要研究方向为图像检测与智能识别. 本文通信作者. E-mail: zcxdsg@163.com

    汪洋:南昌航空大学测试与光电工程学院助教. 主要研究方向为数字图像处理. E-mail: 70876@nchu.edu.cn

Sparse-to-dense Large Displacement Motion Optical Flow Estimation Based on Deep Matching

Funds: Supported by National Natural Science Foundation of China (61866026, 61772255), Advantage Subject Team of Jiangxi Province (20165BCB19007), Outstanding Young Scientist Project of Jiangxi Province (20192BCB23011), and Aeronautical Science Foundation of China (2018ZC56008)
More Information
    Author Bio:

    CHEN Zhen Professor at the School of Measuring and Optical Engineering, Nanchang Hangkong University. His research interest covers image detection and intelligent recognition, and computer vision

    ZHANG Dao-Wen Master student at the School of Measuring and Optical Engineering, Nanchang Hangkong University. His research interest covers image processing and pattern recognition

    ZHANG Cong-Xuan Professor at the School of Measuring and Optical Engineering, Nanchang Hangkong University. His research interest covers image detection and intelligent recognition. Corresponding author of this paper

    WANG Yang Assistant at the School of Measuring and Optical Engineering, Nanchang Hangkong University. His main research interest is digital image processing

  • 摘要: 针对非刚性大位移运动场景的光流计算准确性与鲁棒性问题, 提出一种基于深度匹配的由稀疏到稠密大位移运动光流估计方法. 首先利用深度匹配模型计算图像序列相邻帧的初始稀疏运动场; 其次采用网格化邻域支持优化模型筛选具有较高置信度的图像网格和匹配像素点, 获得鲁棒的稀疏运动场; 然后对稀疏运动场进行边缘保护稠密插值, 并设计全局能量泛函优化求解稠密光流场. 最后分别利用MPI-Sintel和KITTI数据库提供的测试图像集对本文方法和Classic + NL, DeepFlow, EpicFlow以及FlowNetS等变分模型、匹配策略和深度学习光流计算方法进行综合对比与分析, 实验结果表明本文方法相对于其他方法具有更高的光流计算精度, 尤其在非刚性大位移和运动遮挡区域具有更好的鲁棒性与可靠性.
  • 图  1  基于区域划分的深度匹配采样窗口示意图 ((a)参考帧采样窗口; (b)传统匹配方法采样窗口;(c)深度匹配算法采样窗口)

    Fig.  1  Illustration of the deep matching sample window based on the regional division ((a) Sample window of the reference frame; (b) Sample window of the traditional matching method; (c) Sample window of the deep matching method)

    图  2  深度匹配金字塔采样示意图 ((a)第1帧子区域聚合; (b)第2帧子区域聚合)

    Fig.  2  Illustration of the pyramid sampling based deep matching ((a) Subregion polymerization of the first frame; (b) Subregion polymerization of the second frame)

    图  3  本文邻域支持模型运动场优化效果 (蓝色标记符表示匹配正确像素点, 红色标记符表示匹配错误像素点)

    Fig.  3  Optimization effect of the motion field by using the proposed neighborhood supporting model (The blue mark indicates the correct matching pixels, the red mark denotes the false matching pixels)

    图  4  不同参数设置对本文光流估计精度的影响

    Fig.  4  Variation of optical flow estimation results respect to different parameters

    图  5  非刚性大位移与运动遮挡图像序列光流估计结果

    Fig.  5  Optical flow results of the image sequences including non-rigidly large displacements and motion occlusions

    图  6  KITTI数据库测试图像序列光流误差图

    Fig.  6  Optical flow error maps of KITTI dataset

    图  7  MPI-Sintel数据库消融实验光流图

    Fig.  7  Optical flow results of the ablation experiment tested on MPI-Sintel database

    表  1  MPI-Sintel数据库光流估计误差对比

    Table  1  Comparison results of optical flow errors on MPI-Sintel database

    对比方法AAEAEE
    Classic+NL[7]10.125.75
    DeepFlow[18]7.884.12
    EpicFlow[21]8.325.10
    FlowNetS[12]7.554.07
    本文方法7.083.74
    下载: 导出CSV

    表  2  非刚性大位移与运动遮挡图像序列光流估计误差对比

    Table  2  Comparison results of optical flow errors on the image sequences including non-rigidly large displacements and motion occlusions

    对比方法平均误差Ambush_5Cave_2Market_2Market_5Temple_2
    AAE/AEEAAE/AEEAAE/AEEAAE/AEEAAE/AEEAAE/AEE
    Classic+NL[7]14.71/9.2822.53/11.0615.78/14.037.64/0.9818.93/16.598.39/3.72
    DeepFlow[18]10.89/6.6618.86/8.759.23/9.308.00/0.8512.19/11.896.15/2.50
    EpicFlow[21]10.64/6.4719.19/8.487.45/7.817.91/0.8912.15/12.476.48/2.72
    FlowNetS[12]15.63/9.7725.37/12.4317.24./15.668.56/1.2616.56/15.2410.45/4.24
    本文方法9.77/6.1218.43/8.436.98/7.497.05/0.7810.58/11.355.83/2.56
    下载: 导出CSV

    表  3  KITTI数据库光流估计误差对比

    Table  3  Comparison results of optical flow errors on KITTI database

    对比方法AEEnocAEEalloutnoc (%)outall (%)
    Classic+NL[7]11.9812.5913.7820.37
    DeepFlow[18]5.436.139.5114.72
    EpicFlow[21]6.387.5610.9214.18
    FlowNetS[12]239.8285.886.6486.68
    本文方法4.725.058.019.52
    下载: 导出CSV

    表  4  本文方法消融实验结果对比

    Table  4  Comparison results of the ablation experiment

    消融模型Alley_2Cave_4Market_6
    本文方法0.071.163.72
    无匹配优化0.091.285.07
    无稠密插值0.141.315.85
    无全局优化0.091.213.84
    下载: 导出CSV

    表  5  本文方法与其他方法时间消耗对比(s)

    Table  5  Comparison of time consumption between the proposed method and the other approaches (s)

    对比方法MPI-SintelKITTI
    Classic+NL[7]565211
    DeepFlow[18]19.021.2
    EpicFlow[21]16.416.0
    FlowNetS[12]0.101.05
    本文方法88.2104
    下载: 导出CSV
  • [1] 潘超, 刘建国, 李峻林. 昆虫视觉启发的光流复合导航方法. 自动化学报, 2015, 41(6): 1102-1112.

    PAN Chao, LIU Jian-Guo, LI Jun-Lin. An optical flow-based composite navigation method inspired by insect vision. Acta Automatica Sinica, 2015, 41(6): 1102-1112 (in Chinese).
    [2] Colque R, Caetano C, Andrade M, Schwartz W. Histograms of optical flow orientation and magnitude and entropy to detect anomalous events in Videos. IEEE Transactions on Circuits and Systems for Video Technology, 2017, 27(3): 673-682. doi: 10.1109/TCSVT.2016.2637778
    [3] 王飞, 崔金强, 陈本美, 李崇兴. 一套完整的基于视觉光流和激光扫描测距的室内无人机导航系统. 自动化学报, 2013, 39(11): 1889-1900.

    WANG Fei, CUI Jin-Qiang, CHEN Ben-Mei, LEE Tong H. A comprehensive uav indoor navigation system based on vision optical flow and laser fastSLAM. Acta Automatica Sinica, 2013, 39(11): 1889-1900 (in Chinese).
    [4] 张桂梅, 孙晓旭, 刘建新, 储珺. 基于分数阶微分的TV-L1光流模型的图像配准方法研究. 自动化学报, 2017, 43(12): 2213-2224.

    ZHANG Gui-Mei, SUN Xiao-Xu, LIU Jian-Xin, CHU Jun. Research on TV-L1 optical flow model for image registration based on fractional-order differentiation. Acta Automatica Sinica, 2017, 43(12): 2213-2224 (in Chinese).
    [5] Horn B K P, Schunck B G. Determining optical flow. Artificial Intelligence, 1980, 17(1): 185- 203.
    [6] Brox T, Bruhn A, Papenberg N, Weickert J. High accuracy optical flow estimation based on a theory for warping. In: Proceedings of the 2004 European Conference on Computer Vision. Prague, Czech Republic: Springer, 2004. 25−36
    [7] Sun D, Roth S, Black M J. A quantitative analysis of current practices in optical flow estimation and the principles behind them. International Journal of Computer Vision, 2014, 106(2): 115-137. doi: 10.1007/s11263-013-0644-x
    [8] Drulea M, Nedevschi S. Total variation regularization of local-global optical flow. In: Proceedings of the 2011 International Conference on Intelligent Transportation Systems. Washington DC, USA: IEEE, 2011. 318−323
    [9] Perona P, Malik J. Scale-space and edge detection using anisotropic diffusion. IEEE Transactions on pattern analysis and machine intelligence, 1990, 12(7): 629-639. doi: 10.1109/34.56205
    [10] Weickert J, Schnörr C. A theoretical framework for convex regularizers in PDE-based computation of image motion. International Journal of Computer Vision, 2001, 45(3): 245- 264. doi: 10.1023/A:1013614317973
    [11] Zimmer H, Bruhn A, Weickert J. Optic flow in harmony. International Journal of Computer Vision, 2011, 93(3): 368-388. doi: 10.1007/s11263-011-0422-6
    [12] Dosovitskiy A, Fischer P, Ilg E, Häusser P, Hazirbas C, Golkov V, et al. Flownet: Learning optical flow with convolutional networks. In: Proceedings of the 2015 International Conference on Computer Vision and Pattern Recognition. Santiago, Chile: IEEE, 2015. 2758−2766
    [13] Ilg E, Mayer N, Saikia T, Keuper M, Dosovitskiy A, Brox T. FlowNet 2.0: Evolution of optical flow estimation with deep networks. In: Proceedings of the 2017 International Conference on Computer Vision and Pattern Recognition. Honolulu, USA: IEEE, 2017. 1647−1655
    [14] Ranjan A, Black M J. Optical flow estimation using a spatial pyramid network. In: Proceedings of the 2017 International Conference on Computer Vision and Pattern Recognition. Honolulu, USA: IEEE, 2017. 4161−4170
    [15] Hui T W, Tang X, Loy C. LiteFlowNet: A lightweight convolutional neural network for optical flow estimation. In: Proceedings of the 2018 International Conference on Computer Vision and Pattern Recognition. Salt Lake City, USA: IEEE, 2018. 8981−8989
    [16] Ilg E, Saikia T, Keuper M, Brox T. Occlusions, motion and depth boundaries with a generic network for disparity, optical flow or scene flow estimation. In: Proceedings of the 2018 European Conference on Computer Vision. Munich, Germany: Springer, 2018. 614−630
    [17] Brox T, Malik J. Large displacement optical flow: descriptor matching in variational motion estimation. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2011, 33(3): 500-513. doi: 10.1109/TPAMI.2010.143
    [18] Weinzaepfel P, Revaud J, Harchaoui Z, Schmid C. DeepFlow: Large displacement optical flow with deep matching. In: Proceedings of the 2013 International Conference on Computer Vision. Sydney, Australia: IEEE, 2013. 1385−1392
    [19] 张聪炫, 陈震, 熊帆, 黎明, 葛利跃, 陈昊. 非刚性稠密匹配大位移运动光流估计. 电子学报, 2019, 47(6): 1316-1323. doi: 10.3969/j.issn.0372-2112.2019.06.019

    ZHANG Cong-xuan, CHEN Zhen, XIONG Fan, LI Ming, GE Li-yue, CHEN Hao. Large displacement motion optical flow estimation with non-rigid dense patch matching. Acta Electronica Sinica, 2019, 47(6): 1316-1323 (in Chinese). doi: 10.3969/j.issn.0372-2112.2019.06.019
    [20] Hu Y L, Song R, Li Y S. Efficient coarse-to-fine patch match for large displacement optical flow. In: Proceedings of the 2016 International Conference on Computer Vision and Pattern Recognition, Las Vegas, USA: IEEE, 2016. 5704−5712
    [21] Revaud J, Weinzaepfel P, Harchaoui Z, Schmid C. Epicflow: Edge-preserving interpolation of correspondences for optical flow. In: Proceedings of the 2015 International Conference on Computer Vision and Pattern Recognition. Boston, USA: IEEE, 2015. 1164−1172
    [22] Hu Y L, Li Y S, Song R. Robust interpolation of correspondences for large displacement optical flow. In: Proceedings of the 2017 International Conference on Computer Vision and Pattern Recognition, Honolulu, USA: IEEE, 2017. 481−489
    [23] Revaud J, Weinzaepfel P, Harchaoui Z, Schmid C. DeepMatching: Hierarchical deformable dense matching. International Journal of Computer Vision, 2016, 120(3): 300-323. doi: 10.1007/s11263-016-0908-3
    [24] Bian J W, Lin W Y, Matsushita Y, Yeung S K, Nguyen T D, Cheng M M. GMS: Grid-based motion statistics for fast, ultra- robust feature correspondence. In: Proceedings of the 2017 International Conference on Computer Vision and Pattern Recognition. Honolulu, USA: IEEE, 2017. 4181−4190
  • 加载中
图(7) / 表(5)
计量
  • 文章访问数:  487
  • HTML全文浏览量:  435
  • PDF下载量:  130
  • 被引次数: 0
出版历程
  • 收稿日期:  2019-10-16
  • 录用日期:  2020-04-06
  • 网络出版日期:  2022-08-10
  • 刊出日期:  2022-09-16

目录

    /

    返回文章
    返回