• Laser & Optoelectronics Progress
  • Vol. 58, Issue 20, 2010006 (2021)
Haidong Zhang, Yiming Xu*, Li Wang, Chunlei Bian, and Fangjie Zhou
Author Affiliations
  • School of Electrical Engineering, Nantong University, Nantong, Jiangsu 226019, China
  • show less
    DOI: 10.3788/LOP202158.2010006 Cite this Article Set citation alerts
    Haidong Zhang, Yiming Xu, Li Wang, Chunlei Bian, Fangjie Zhou. Visual Odometry Based on Improved Dual-Stream Network Structure[J]. Laser & Optoelectronics Progress, 2021, 58(20): 2010006 Copy Citation Text show less
    System overall structure
    Fig. 1. System overall structure
    Network framework
    Fig. 2. Network framework
    Inception structure. (a) Incep1 structure; (b) Incep2 structure
    Fig. 3. Inception structure. (a) Incep1 structure; (b) Incep2 structure
    CBAM structure diagram
    Fig. 4. CBAM structure diagram
    Processed data set. (a) RGB image; (b) depth image
    Fig. 5. Processed data set. (a) RGB image; (b) depth image
    Ground truth processing
    Fig. 6. Ground truth processing
    Diagram of partly sequence 00--10 trajectory
    Fig. 7. Diagram of partly sequence 00--10 trajectory
    Image acquisition platform
    Fig. 8. Image acquisition platform
    Experimental scene and trajectory diagram
    Fig. 9. Experimental scene and trajectory diagram
    LayerFilter sizeStridePaddingOutput data dimension
    WidthHeightDepth (number of channels)
    Input4161286
    Incep1(Conv1)Multiple (7×7)232086416
    Incep2(Conv2)Multiple (5×5)221043232
    Attention modelMultiple1043232
    Conv33×321521664
    Conv43×321268128
    Conv53×321134256
    Conv63×32172256
    Conv73×32141512
    Flatten2048
    Concat4096
    FC1512
    FC264
    FC36
    Table 1. Network layer parameters
    SequenceVISO2-MSfMLearnerVISO2-SProposed method
    trmse/%rrmse/[102(°)·m-1]trmse/%rrmse/[102(°)·m-1]trmse/%rrmse/[102(°)·m-1]trmse/%rrmse/[102(°)·m-1]
    0012.053.6723.546.191.860.535.022.11
    0125.557.3959.312.7923.900.312.290.72
    0316.641.3411.974.182.100.423.191.24
    0422.861.6921.133.282.100.271.681.23
    0516.294.5817.144.661.500.502.861.36
    0611.191.9914.875.581.540.331.020.50
    0738.079.2020.986.311.900.932.971.95
    0833.562.4817.413.751.960.605.822.36
    0913.621.7211.114.071.850.463.381.26
    1066.118.0823.304.061.280.534.921.94
    Mean25.594.2122.084.491.790.513.321.47
    Table 2. Test results of VISO-M, SfMLearner,VISO-S, and proposed method on the KITTI Visual Odometry data set
    Haidong Zhang, Yiming Xu, Li Wang, Chunlei Bian, Fangjie Zhou. Visual Odometry Based on Improved Dual-Stream Network Structure[J]. Laser & Optoelectronics Progress, 2021, 58(20): 2010006
    Download Citation