• Chinese Journal of Lasers
  • Vol. 49, Issue 23, 2310003 (2022)
Siqi Wang1、3, Jiaqiang Zhang1、3, Liyuan Li1、3、*, Xiaoyan Li1、2, and Fansheng Chen1、2
Author Affiliations
  • 1Key Laboratory of Intelligent Infrared Perception, Shanghai Institute of Technical Physics, Chinese Academy of Sciences, Shanghai 200083, China
  • 2Hangzhou Institute for Advanced Study, University of Chinese Academy of Sciences, Hangzhou 310024, Zhejiang, China
  • 3University of Chinese Academy of Sciences, Beijing 100049, China
  • show less
    DOI: 10.3788/CJL202249.2310003 Cite this Article Set citation alerts
    Siqi Wang, Jiaqiang Zhang, Liyuan Li, Xiaoyan Li, Fansheng Chen. Application of MVSNet in 3D Reconstruction of Space Objects[J]. Chinese Journal of Lasers, 2022, 49(23): 2310003 Copy Citation Text show less
    Structural diagram of MVSNet network
    Fig. 1. Structural diagram of MVSNet network
    Architecture diagram of feature extraction network
    Fig. 2. Architecture diagram of feature extraction network
    Architecture diagram of cost volume regularization network
    Fig. 3. Architecture diagram of cost volume regularization network
    Camera location map
    Fig. 4. Camera location map
    Space target dataset
    Fig. 5. Space target dataset
    Reconstruction results for non-Lambertian reflector region of Clementine satellite obtained by different methods. (a)Ground truth;(b)VisualSFM;(c)COLMAP;(d)SurfaceNet;(e) our method
    Fig. 6. Reconstruction results for non-Lambertian reflector region of Clementine satellite obtained by different methods. (a)Ground truth;(b)VisualSFM;(c)COLMAP;(d)SurfaceNet;(e) our method
    Reconstruction results for low-textured region of Clementine satellite obtained by different methods. (a)Ground truth;(b)VisualSFM;(c)COLMAP;(d)SurfaceNet;(e) our method
    Fig. 7. Reconstruction results for low-textured region of Clementine satellite obtained by different methods. (a)Ground truth;(b)VisualSFM;(c)COLMAP;(d)SurfaceNet;(e) our method
    Reconstruction results for repetitive texture region of Clementine satellite obtained by different methods. (a)Ground truth;(b)VisualSFM;(c)COLMAP;(d)SurfaceNet;(e) our method
    Fig. 8. Reconstruction results for repetitive texture region of Clementine satellite obtained by different methods. (a)Ground truth;(b)VisualSFM;(c)COLMAP;(d)SurfaceNet;(e) our method
    MethodMean accuracy error /mmMean completeness error /mmOverall error /mm
    Our method0.4490.3790.414
    SurfaceNet0.5100.7480.629
    COLMAP0.4000.6440.522
    VisualSFM0.6130.8350.724
    Table 1. Performance comparison of different algorithms on DTU dataset
    MethodOur methodSurfaceNetCOLMAPVisualSFM
    Running time /s230368022700223
    Table 2. Running time
    Number of matching viewsMean accuracy error /mmMean completeness error /mmOverall error /mm
    20.4770.4310.454
    30.4490.3790.414
    40.4320.3540.393
    50.4290.3450.387
    Table 3. Performance comparison of algorithms under different numbers of matching views
    Siqi Wang, Jiaqiang Zhang, Liyuan Li, Xiaoyan Li, Fansheng Chen. Application of MVSNet in 3D Reconstruction of Space Objects[J]. Chinese Journal of Lasers, 2022, 49(23): 2310003
    Download Citation