richardson lucy deblurring for moving light field cameras
play

Richardson-Lucy Deblurring for Moving Light Field Cameras Donald - PowerPoint PPT Presentation

Richardson-Lucy Deblurring for Moving Light Field Cameras Donald Dansereau 1 , Anders Eriksson 2 and Jrgen Leitner 2,3 1 Stanford University, 2 Queensland University of Technology, 3 ARC Centre of Excellence for Robotic Vision CVPR:LF4CV


  1. Richardson-Lucy Deblurring for Moving Light Field Cameras Donald Dansereau 1 , Anders Eriksson 2 and Jürgen Leitner 2,3 1 Stanford University, 2 Queensland University of Technology, 3 ARC Centre of Excellence for Robotic Vision CVPR:LF4CV Workshop 2017 July 26

  2. 3D Motion Complicates Vision Scene-dependent nonuniform apparent motion 2 ComputationalImaging.org

  3. 3D is Easier in 4D We have 6-DOF virtual camera control Video Stabilization [video] http://pages.cs.wisc.edu/~lizhang/projects/lfstable/ [Smith2009] 3 ComputationalImaging.org

  4. 3D is Easier in 4D We can fix the camera’s position Closed-Form Change Detection Per-pix ixel l stil still- l-ca camera meth thods ● Change detection ● Tracking/segmentation ● Velocity & temporal filtering [dansereau2016] http://dgd.vision/Projects/LFChangeDet/ 4 ComputationalImaging.org

  5. 3D is Easier in 4D We can fix the camera’s position Closed-Form Change Detection Per-pix ixel l stil still- l-ca camera meth thods ● Change detection ● Tracking/segmentation ● Velocity & temporal filtering [dansereau2016] http://dgd.vision/Projects/LFChangeDet/ 5 ComputationalImaging.org

  6. 3D is Easier in 4D We can fix the camera’s position Closed-Form Change Detection Per-pix ixel l stil still- l-ca camera meth thods ● Change detection ● Tracking/segmentation ● Velocity & temporal filtering [dansereau2016] http://dgd.vision/Projects/LFChangeDet/ 6 ComputationalImaging.org

  7. 3D is Easier in 4D Lukas-Kanade optical flow generalizes to 6-DOF Linearize Apparent Motion Closed-form 6-DOF Odometry L x L y T r u e P l e n o p t i c P o i n t w i s e F e a t u r e 2 ) m ( − 2 z L y L x 1 0 1 4 1 2 5 1 0 0 8 − 5 6 4 − 1 0 y ( m ) x ( m ) 2 − 1 5 L z L z [Neumann2002, Dansereau2011, Dong2013] 7 ComputationalImaging.org

  8. 3D is Easier in 4D Lukas-Kanade optical flow generalizes to 6-DOF Linearize Apparent Motion Closed-form 6-DOF Odometry L x L y T r u e P l e n o p t i c P o i n t w i s e F e a t u r e 2 ) m ( − 2 z L y L x 1 0 1 4 1 2 5 1 0 0 8 − 5 6 4 − 1 0 y ( m ) x ( m ) 2 − 1 5 L z L z [Neumann2002, Dansereau2011, Dong2013] 8 ComputationalImaging.org

  9. Blur in 3D Scenes Convolution models blurring in 2D… Can we replace convolution with LF rendering in 3D scenes? 9 ComputationalImaging.org

  10. Related Work “Light Field Blind Motion Deblurring” [Srinivasan 2017] ● 3-DOF ● Insights on blur manifestation in LF ● Blind ● Modern optimization (ADAM) LF-RL ● Requires extension to be blind ● 6-DOF ● Proof of convergence to ML estimate (see paper) ● New LF equiparallax regularizer 10 ComputationalImaging.org

  11. Related Work “Richardson-Lucy Deblurring for Scenes under a Projective Motion Path” [Tai et al. 2011] 11 ComputationalImaging.org

  12. Richardson-Lucy Deblurring IN 12 ComputationalImaging.org

  13. Light Field Richardson-Lucy IN FWD BLUR REV BLUR Still LF 13 ComputationalImaging.org

  14. Light Field Richardson-Lucy IN FWD BLUR REV BLUR Still LF Simulated Motion 14 ComputationalImaging.org

  15. Light Field Richardson-Lucy IN FWD BLUR REV BLUR Still LF Simulated Motion Simulated Blur 15 ComputationalImaging.org

  16. Regularization Anisotropic total variation Favour textural edges [Goldluecke & Wanner 2013, Heber2013] Equiparallax Favour equal slopes in s,u and t,v 16 ComputationalImaging.org

  17. Rendered Results: Rot about y Ori riginal Ori riginal Ori riginal 17 ComputationalImaging.org

  18. Rendered Results: Rot about y Blurre Ori riginal red Ori Bl Blurre riginal red Ori Bl Blurre riginal red 18 ComputationalImaging.org

  19. Rendered Results: Rot about y Deblurre Blurre Ori riginal red rred Deblurr Ori Bl Blurre riginal red rred Deblurr Blurre Bl Ori riginal red rred 19 ComputationalImaging.org

  20. Rendered Results: Rot about z rotz Bl Blurre rred Bl Blurr rred Bl Blurr rred 20 ComputationalImaging.org

  21. Rendered Results: Rot about z rotz Bl Blurre Deblurre rred red Blurr Deblurre Bl rred rred Deblurre Bl Blurr rred rred 21 ComputationalImaging.org

  22. Rendered Results: Trans along x Bl Blurre rred Bl Blurr rred Bl Blurr rred 22 ComputationalImaging.org

  23. Rendered Results: Trans along x Bl Blurre Deblurre rred red Blurr Deblurre Bl rred rred Deblurre Blurr Bl rred rred 23 ComputationalImaging.org

  24. Rendered Results: Rot about z rotz Bl Blurre rred Bl Blurr rred Bl Blurr rred 24 ComputationalImaging.org

  25. Rendered Results: Rot about z rotz Bl Blurre Deblurre rred red Blurr Deblurre Bl rred rred Deblurre Bl Blurr rred rred 25 ComputationalImaging.org

  26. Rendered Results: Trans along z Bl Blurre rred Blurr Bl rred Bl Blurr rred 26 ComputationalImaging.org

  27. Rendered Results: Trans along z Deblurre Bl Blurre rred red Bl Blurr Deblurre rred rred Bl Blurr Deblurre rred rred 27 ComputationalImaging.org

  28. Rendered Results: Trans along z Deblurre Blurre Bl rred red Without Regulariza Bl Blurr Deblurre rred rred zation Without Regulariza Deblurre Blurr Bl rred rred zation 28 ComputationalImaging.org

  29. Results: Captured Quantitative Evaluation Repeatable camera motion Isolated dimensions Known magnitudes George 29 ComputationalImaging.org

  30. Validating Calibration & Rendering Still Camera calibration, rectification Metric blur simulation Metric robot motion Camera-to-robot calibration 30 ComputationalImaging.org

  31. Validating Calibration & Rendering Measured Still Blur Camera calibration, rectification Metric blur simulation Metric robot motion Camera-to-robot calibration 31 ComputationalImaging.org

  32. Validating Calibration & Rendering Simulated Blur Measured Still Blur Camera calibration, rectification Metric blur simulation Metric robot motion Camera-to-robot calibration 32 ComputationalImaging.org

  33. Validating Calibration & Rendering Simulated Blur Measured Still Still Blur Camera calibration, rectification Metric blur simulation Metric robot motion Camera-to-robot calibration 33 ComputationalImaging.org

  34. Results: Captured No increase in noise Large increase in sharpness Regularization is helping 34 ComputationalImaging.org

  35. Results: Captured Measured Blur 35 ComputationalImaging.org

  36. Results: Captured Measured Blur Deblurred 36 ComputationalImaging.org

  37. Results: Captured 37 ComputationalImaging.org

  38. Results: Captured 38 ComputationalImaging.org

  39. Results: Captured 39 ComputationalImaging.org

  40. Results: Captured 40 ComputationalImaging.org

  41. Results: Captured 41 ComputationalImaging.org

  42. Results: Captured 42 ComputationalImaging.org

  43. Summary & Future Work Generaliz lized co convolu lutio tional l blu lur usin sing LF Renderin ing Applied to RL deblurring 3D scenes, 6-DOF camera motion Proof of convergence to ML estimate Equiparallax regularization Next: Equiparallax regularization: applications Beyond 6-DOF, defocus Blind deblurring 43 ComputationalImaging.org

  44. Acknowledgments QUT HPC Group George 44 ComputationalImaging.org

  45. Light Field T oolbox for MATLAB Load Gantry and Lytro imagery Calibrate and rectify Lytro imagery Linear depth, volume fjlters Denoising: low-light, fog, dust, murky water Occluder removal: rain, snow, silty water LF Synth: Bare-Bones Rendering 45 ComputationalImaging.org

Recommend


More recommend