• Acta Optica Sinica
  • Vol. 44, Issue 18, 1812001 (2024)
Xiaoyu Gu, Xiaojing Gu*, Jie Ding, and Xingsheng Gu
Author Affiliations
  • Key Laboratory of Smart Manufacturing in Energy Chemical Process, Ministry of Education, East China University of Science and Technology, Shanghai 200237, China
  • show less
    DOI: 10.3788/AOS240598 Cite this Article Set citation alerts
    Xiaoyu Gu, Xiaojing Gu, Jie Ding, Xingsheng Gu. Estimation of Gas Velocity in Optical Gas Imaging Based on Deep Optical Flow Network[J]. Acta Optica Sinica, 2024, 44(18): 1812001 Copy Citation Text show less
    Framework for generating gas optical flow dataset
    Fig. 1. Framework for generating gas optical flow dataset
    Schematic diagram of ray marching method
    Fig. 2. Schematic diagram of ray marching method
    Three-layer radiative transfer model
    Fig. 3. Three-layer radiative transfer model
    Synthesized gas optical flow dataset. (a)(d)(g) Frame 1; (b)(e)(h) frame 2; (c)(f)(i) optical flow label
    Fig. 4. Synthesized gas optical flow dataset. (a)(d)(g) Frame 1; (b)(e)(h) frame 2; (c)(f)(i) optical flow label
    Schematic diagram of gas column density and velocity estimation
    Fig. 5. Schematic diagram of gas column density and velocity estimation
    Estimation results of multiple optical flow methods on the gas optical flow validation set. (a) Frame 1; (b) frame 2; (c) optical flow label; (d) Farnebäck; (e) DIS; (f) FlowNet2; (g) PWC-Net; (h) RAFT; (i) GMA; (j) FlowNet2+ours
    Fig. 6. Estimation results of multiple optical flow methods on the gas optical flow validation set. (a) Frame 1; (b) frame 2; (c) optical flow label; (d) Farnebäck; (e) DIS; (f) FlowNet2; (g) PWC-Net; (h) RAFT; (i) GMA; (j) FlowNet2+ours
    Estimated optical flow results of real images. (a) Frame 1; (b) frame 2; (c) Farnebäck; (d) DIS; (e) FlowNet2; (f) PWC-Net; (g) RAFT; (h) GMA; (i) ours
    Fig. 7. Estimated optical flow results of real images. (a) Frame 1; (b) frame 2; (c) Farnebäck; (d) DIS; (e) FlowNet2; (f) PWC-Net; (g) RAFT; (h) GMA; (i) ours
    Continuous synthetic gas optical flow data. (a)(b)(c)(d) Frame 1, 20, 40, 60 gas images; (e)(f)(g)(h) frame 1, 20, 40, 60 gas optical flow label
    Fig. 8. Continuous synthetic gas optical flow data. (a)(b)(c)(d) Frame 1, 20, 40, 60 gas images; (e)(f)(g)(h) frame 1, 20, 40, 60 gas optical flow label
    Comparison of warping transformations between the FlyingChairs and gas optical flow dataset. (a) Origin image in FlyingChairs; (b) origin image in gas optical flow dataset; (c) warped image in FlyingChairs; (d) warped image in gas optical flow dataset
    Fig. 9. Comparison of warping transformations between the FlyingChairs and gas optical flow dataset. (a) Origin image in FlyingChairs; (b) origin image in gas optical flow dataset; (c) warped image in FlyingChairs; (d) warped image in gas optical flow dataset
    Comparisons of optical flow estimation results using different loss functions. (a)(b) Frame 1; (c)(d) frame 2; (e)(f) optical flow label; (g)(h) optical flow estimation results with gradient loss; (i)(j) optical flow estimation results without gradient loss
    Fig. 10. Comparisons of optical flow estimation results using different loss functions. (a)(b) Frame 1; (c)(d) frame 2; (e)(f) optical flow label; (g)(h) optical flow estimation results with gradient loss; (i)(j) optical flow estimation results without gradient loss
    MethodAEPEAAE
    Farnebäck74.481.26
    DIS303.950.90
    FlowNet2144.030.92
    PWC-Net154.150.34
    RAFT164.460.87
    GMA174.541.18
    FlowNet2+Rangels183.990.91
    PWC-Net+ours1.730.34
    RAFT+ours1.290.31
    GMA+ours1.350.30
    FlowNet2+ours1.270.27
    Table 1. Performance comparison of different optical flow methods on the gas optical flow validation dataset
    MethodAccuracy /%
    Farnebäck723.66
    DIS3072.90
    FlowNet21459.64
    PWC-Net1538.81
    RAFT1672.59
    GMA1760.93
    FlowNet2+Rangels1865.51
    FlowNet2+ours81.66
    Table 2. Comparison of velocity estimation accuracy of different optical flow methods on continuous synthetic gas data
    ConditionFlowNet214PWC-Net15RAFT16GMA17
    4.034.154.464.54
    +finetune4.745.064.514.59
    +finetune+background1.291.731.311.37
    +finetune +gradientloss5.094.884.524.59
    +finetune+background+gradientloss1.271.731.291.35
    Table 3. AEPE obtained from different methods in ablation study
    Xiaoyu Gu, Xiaojing Gu, Jie Ding, Xingsheng Gu. Estimation of Gas Velocity in Optical Gas Imaging Based on Deep Optical Flow Network[J]. Acta Optica Sinica, 2024, 44(18): 1812001
    Download Citation