• Laser & Optoelectronics Progress
  • Vol. 56, Issue 11, 111505 (2019)
Peng Xiang, Bin Zhou*, Yangkun Zhu, Wenkai He, Xiaogeng Yue, and Yibei Tao
Author Affiliations
  • School of Energy and Environment, Southeast University, Nanjing, Jiangsu 210096, China
  • show less
    DOI: 10.3788/LOP56.111505 Cite this Article Set citation alerts
    Peng Xiang, Bin Zhou, Yangkun Zhu, Wenkai He, Xiaogeng Yue, Yibei Tao. Camera Calibration Based on Deep Neural Network in Complex Environments[J]. Laser & Optoelectronics Progress, 2019, 56(11): 111505 Copy Citation Text show less
    Model of shallow neural network
    Fig. 1. Model of shallow neural network
    Framework of DNN
    Fig. 2. Framework of DNN
    Loss function convergence curves under different parameter update algorithms
    Fig. 3. Loss function convergence curves under different parameter update algorithms
    Loss function convergence curves under different learning rates
    Fig. 4. Loss function convergence curves under different learning rates
    Twenty-five trials for random search
    Fig. 5. Twenty-five trials for random search
    Loss and training time of test points after iteration
    Fig. 6. Loss and training time of test points after iteration
    Calibration target
    Fig. 7. Calibration target
    Highly distorted image
    Fig. 8. Highly distorted image
    Angle /(°)Error of DNN method /mmError of SNN method /mmError of Zhang's method / mm
    AvgMaxStdAvgMaxStdAvgMaxStd
    00.08960.41090.05531.89108.49891.38400.38872.26140.2714
    150.09480.57740.06573.00279.12131.67060.47681.71190.3339
    300.12710.71460.08953.379311.88042.14580.37223.03250.2733
    450.14920.78310.09303.988915.56272.62220.25861.77890.2139
    600.18970.89290.10034.651517.30862.82260.53503.28830.4284
    Table 1. Calibration errors of different methods under different shooting angles
    ErrorDNN methodSNN methodZhang's method
    Avg0.14717.88962.8714
    Max0.899034.26697.1937
    Std0.11244.86281.7055
    Table 2. Calibration errors of different methods with highly distorted lensmm
    ErrorWithhomogeneous termWithouthomogeneous term
    Avg0.09480.1325
    Max0.57740.8654
    Std0.06570.0992
    Table 3. Calibration errors of DNN with different structuresmm
    Peng Xiang, Bin Zhou, Yangkun Zhu, Wenkai He, Xiaogeng Yue, Yibei Tao. Camera Calibration Based on Deep Neural Network in Complex Environments[J]. Laser & Optoelectronics Progress, 2019, 56(11): 111505
    Download Citation