Chinese Journal of Liquid Crystals and Displays, Volume. 38, Issue 10, 1434(2023)

Optical flow estimation via fusing sequence image intensity correlation information

Tong AN1, Di JIA1,2、*, Jia-bao ZHANG1, and Peng CAI1
Author Affiliations
  • 1College of Electronic and Information Engineering,Liaoning Technical University,Huludao 125105,China
  • 2College of Electrical and Control Engineering,Liaoning Technical University,Huludao 125105,China
  • show less
    Figures & Tables(12)
    Optical flow estimation network structure
    Encoder structure
    Strength-weakness correlation of the edge-region point mapping with the center-region point mapping done by the horizontal cut(The strength of the correlation is indicated by the shade of the color of the bars,The correlation factor is set to 1 when the corresponding position area is identified as strong correlation. The weaker the rest of the color,the weaker the correlation).
    Lookup operator based on correlation pyramid. The orange part is the neighborhood dot product of the query points at different scales.
    Basic motion encoder module structure design
    Convergence curve on KITTI-2015.(a)Convergence curve on EPE index;(b)Convergence curve on Fl index.
    Optical flow estimation results on the KITTI validation set(4,6,and 8 are the 4-point,6-point,and 8-point methods proposed in this paper,respectively).
    Convergence curve on MPI-Sintel.(a)Convergence curve on EPE index;(b)Convergence curve on 1 px index;(c)Convergence curve on 3 px index;(d)Convergence curve on 5 px index.
    Optical flow estimation results on the MPI- Sintel validation set(4,6,and 8 are the 4-point,6-point,and 8-point methods proposed in this paper,respectively).
    • Table 1. Optical flow estimation performance of different methods on KITTI-2015 test set(↓:The smaller the value,the better the performance)

      View table
      View in Article

      Table 1. Optical flow estimation performance of different methods on KITTI-2015 test set(↓:The smaller the value,the better the performance)

      方法EPE↓Fl↓
      VCN131.417 4022.675 603
      DICL211.319 6552.679 850
      RAFT180.769 7992.165 965
      Ours(4)0.706 6511.850 664
      Ours(6)0.712 9631.849 031
      Ours(8)0.707 3471.862 571
    • Table 2. Optical flow estimation performance of different methods on MPI-Sintel test set(↓:The smaller the value,the better the performance;↑:The greater the value,the better the performance)

      View table
      View in Article

      Table 2. Optical flow estimation performance of different methods on MPI-Sintel test set(↓:The smaller the value,the better the performance;↑:The greater the value,the better the performance)

      方法CleanFinal
      EPE↓1 px↑3 px↑5 px↑EPE↓1 px↑3 px↑5 px↑
      VCN131.291 9130.829 4500.879 8560.896 3901.768 7390.812 3060.882 6260.903 631
      DICL210.974 6820.883 8690.935 7150.949 7041.382 8320.857 1640.925 6160.945 201
      RAFT180.891 9260.901 7240.959 2830.972 9311.278 3160.869 3130.938 7210.958 746
      Ours(4)0.840 8110.910 8940.962 3910.974 7161.203 1680.878 0460.942 5200.961 372
      Ours(6)0.847 9120.911 2200.962 4030.974 7911.206 4910.878 1680.942 8540.960 920
      Ours(8)0.837 0720.910 8540.962 2890.974 7441.239 3960.877 5900.942 1260.960 501
    • Table 3. Ablation experiments on KITTI-2015 and MPI-Sintel datasets

      View table
      View in Article

      Table 3. Ablation experiments on KITTI-2015 and MPI-Sintel datasets

      KITTI-2015MPI-Sintel
      EPEFlClean(EPE)Clean(1 px)Clean(3 px)Clean(5 px)Final(EPE)Final(1 px)Final(3 px)Final(5 px)
      BC0.718 101.880 420.848 480.900 940.962 390.964 671.217 430.868 730.942 350.960 99
      BA(4)0.747 232.056 190.864 820.907 660.961 340.974 411.289 550.874 920.941 000.959 64
      CA(4)0.744 561.976 670.873 710.905 160.960 560.973 711.226 330.873 510.940 510.959 65
      BA(6)0.746 182.097 800.865 450.908 040.961 140.974 141.280 900.874 180.940 800.959 46
      CA(6)0.749 142.034 940.870 990.906 350.961 490.974 541.240 370.874 620.941 630.960 46
      BA(8)0.748 252.122 270.865 350.907 420.961 200.974 131.278 820.875 240.941 340.941 34
      CA(8)0.750 021.997 000.870 240.906 370.961 290.974 331.251 650.874 510.941 020.941 34
      BCA(8)0.707 341.850 660.837 070.910 850.962 280.974 741.239 390.877 590.942 120.960 50
    Tools

    Get Citation

    Copy Citation Text

    Tong AN, Di JIA, Jia-bao ZHANG, Peng CAI. Optical flow estimation via fusing sequence image intensity correlation information[J]. Chinese Journal of Liquid Crystals and Displays, 2023, 38(10): 1434

    Download Citation

    EndNote(RIS)BibTexPlain Text
    Save article for my favorites
    Paper Information

    Category: Research Articles

    Received: Nov. 18, 2022

    Accepted: --

    Published Online: Oct. 25, 2023

    The Author Email: Di JIA (1319423118@qq.com)

    DOI:10.37188/CJLCD.2022-0384

    Topics