1 / 53

How to fit a surface to a point cloud? or optimization of surface functionals in computer vision

TRICS seminar Computer Science Department. How to fit a surface to a point cloud? or optimization of surface functionals in computer vision Yuri Boykov. Optimization of surface functionals in computer vision. Computer vs. human vision Model fitting in computer vision

terra
Download Presentation

How to fit a surface to a point cloud? or optimization of surface functionals in computer vision

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. TRICS seminar Computer Science Department How to fit a surface to a point cloud? or optimization of surface functionals in computer vision Yuri Boykov

  2. Optimization of surface functionals in computer vision • Computer vs. human vision • Model fitting in computer vision • templates, pictorial structures, trees, deformable models, contours/snakes, meshes, surfaces, complexes, graphs, weak-membrane model, Mumford-Shah, Potts model,…… • Optimization in computer vision • dynamic programming, gradient descent, PDEs, shortest paths, min. spanning trees, linear and quadratic programming, primal-dual schema, network flow algorithms, QPBO, ... • Applications • segmentation, stereo, multi-view reconstruction, optical flows • surface fitting

  3. Contours

  4. +Shading 3D shape understanding M.C. Escher Drawing hands

  5. +Color Recognition Da Vinci Madonna Litta

  6. +Texture recognizing material Magritte Souvenir de Voyage

  7. +Texture recognizing 3D perspective The New Yorker Album of Drawings, The Viking Press, NY, 1975

  8. What do humans get by ‘looking’? basic image cues: • Contours • Shading • Color • Texture • … J. Vermeer The Guitar Player

  9. What do humans get by ‘looking’? basic image cues: higher-level perception: • Segmentation • Motion • 3D shape perception • 3D scene geometry • Detection/Recognition • … Contours Shading Color Texture …

  10. What do computers get by ‘looking’? x I(x,y) x y y y 3D plot of image intensity I(x,y) x

  11. What do computers get by ‘looking’? higher-level grouping? basic image cues: • Intensity discontinuities(contours) • Intensity gradients (shading) • Multi-valued intensities (color) • Filtering (e.g. texture) • … P. Picasso The Guitar Player

  12. Bayesian approach Prior + Data Fit some prior model into data model Low-level cues (local info) high-level knowledge (global picture)

  13. Face template image Rigid Template Matching image translation, rotation, scaling • In matching we estimate “position” of a rigid template in the image • “Position” includes global location parameters of a rigid template: • - translation, rotation, scale,…

  14. Non-rigid (parametric) matching panorama mosaicing • Pick one image (red) • Warpthe other images to match it (homographic transform) • Blend

  15. e.g.… using homographies

  16. e.g…. using flexible templates • In flexible template matching we estimate “position” of each rigid component of a template • For tree-structured models, efficient global optimization is possible via DP (Felzenswalb&Huttenlocher 2002)

  17. tracking parameters => activity recognition Bottom-up tracker

  18. initial final intermediate deformable contours (“snakes”) Kass, Witkin, Terzopoulos 1987 • 2D curve which matches to image data • Initialized near target, iteratively refined • Can restore missing data Optimization gets harder when a loop is introduced. DP does not apply. One solution: gradient descent

  19. local minima, fixed contour topology Cremers, Tischhäuser, Weickert, Schnörr, “Diffusion Snakes”, IJCV '02

  20. Level set function u(x,y) is normally discretized/stored over image pixels • Values of u(p) can be interpreted asdistancesorheights of image pixels -0.8 -0.8 -1.7 0.2 A contour may be approximated from u(x,y) with near sub-pixel accuracy -0.5 0.5 -0.6 -0.4 0.7 -0.2 0.3 0.6 Implicit representation of contours Osher&Sethian 1989 C

  21. [Visualization is courtesy of O. Juan]Simple evolution Morphological Operation: Erosion

  22. Visualization is courtesy of O. JuanExample of gradient descent evolution Gradient descent w.r.t. Euclideanlength

  23. Example of gradient descent evolution Gradient descent w.r.t. Euclideanlength Laplacian Osher&Sethian 1989

  24. [example from Goldenberg, Kimmel, Rivlin, Rudzsky, IEEE TIP ’01]Geodesic Active Contours via Level-sets

  25. W Other geometric energy functionals besides length[courtesy of Ron Kimmel] Geometric measures commonly used in segmentation gradient descent evolution Functional E( C ) weighted length weighted area alignment (flux)

  26. in 3D…deformable meshes, level-sets, … Estimation of position for mesh points Many loops. optimization - gradient descent Typical problems: - local minima (clutter, outliers) -over-smoothing GOALS: global optima (?) “right” functional (?)

  27. Global Optimization and Surface Functionals

  28. More generally...Estimate labels for graph nodes observed noisy image I image labeling L (restored intensities) I L NOTE: similar to robust regression model estimation p along one scan line in the image

  29. (simple example) Piece-wise smooth restoration • Markov Random Fields (MRF) approach weak membrane model(Geman&Geman’84, Blake&Zisserman’83,87) (Continuous analogue: Mamford-Shah functional, 1989) discontinuity preserving prior optimizing E(L) is NP hard!

  30. (simple example) Piece-wise constant restoration observed noisy image I image labeling L (restored intensities) I L p along one scan line in the image

  31. (simple example) Piece-wise constant restoration Potts model Boykov Veksler Zabih ’01 Greiget al.’89 (for 2 labels) “perceptual grouping” global optimization is still NP hard, but there are fast provably good combinatorial approximation algorithms, linear and quadratic programming, QPBO, primal-dual schema

  32. Perceptual grouping from stereo constant label = plane (Birchfield &Tomasi’99)

  33. Binary labeling(binary image restoration) Greig Porteous Seheult ’89 optimal binary labelingL original binary image I • Globally optimal solution is possible using combinatorial graph cut algorithms • pseudo-boolean optimization Hammer’65, Picard&Ratlif’75

  34. Binary labeling(object extraction) object segmentation left ventricle of heart

  35. Boykov&Jolly’01 Binary labeling(object extraction) surface extraction C left ventricle of heart • Globally optimal solution is possible using graph cut algorithms • pseudo-boolean optimization(Hammer’65, Picard&Ratlif’75)

  36. 0 0 0 1 0 1 0 0 1 0 1 1 Implicitsurface representation via graph-cuts • Any contour (or surface in 3D) satisfying labeling of exterior/interior points (pixel centers) is acceptable if some explicit surface has to be output.

  37. Tight characterization for geometric functionals of contour C that can be globally optimized by graph cut algorithms (Kolmogorov&Boykov’05) disclaimer: for pairwise interactions only Global optimization of geometric surface functionals “edge alignment” • Regional bias • any scalar function f • Geometric length • any convex, • symmetric metric • (e.g. Riemannian) • Flux • any vector field v

  38. Globally optimal surfaces in 3D Volumetric segmentation (BJ01,BK’03,KB’05)

  39. Binary labeling(object extraction) Blake et al.’04, Rother et al.’04 iteratively re-estimate color models e.g. using mixture of Gaussians

  40. Segmentation for Image Blending

  41. Segmentation for Image Blending

  42. Optimal surfaces in 3D Local cues: voxel’sphotoconsistency Prior: smoothness, projective geometry constraints 3D reconstruction Vogiatzis, Torr, Cippola’05

  43. Globally optimal surfaces in 3D from a cheapdigital camera Lempitsky&Boykov, 2006

  44. Globally optimal surfaces in 3D 3D model (texture mapped) multi-view reconstruction set up Furukawa&Ponce2006

  45. 3D model: Surface fitting to point cloud a cloud of 3D points (e.g. from a laser scanner) surface fitting:

  46. Surface fitting to point cloud

  47. Surface fitting functional Fitting a surface into a cloud of oriented points (Lempitsky&Boykov, 2007) data fit prior

  48. From 10 views No initialization is needed Optimal surfaces in 3D Fitting a surface into a cloud of oriented points (Lempitsky&Boykov, 2007)

  49. Global vs. local optimization Fitting a surface into a cloud of oriented points (Lempitsky&Boykov, 2007) initial solution regional potentials global minima local minima

  50. Fitting to sparse data

More Related