1 / 27

Stereo Vision using PatchMatch Algorithm

Stereo Vision using PatchMatch Algorithm. Junkyung Kim Class of 2014. 1. Introduction. Vision Problem Revisited. Loss of information 3-D physical world projected onto a 2-D surface. Vision Problem Revisited. Therefore, an inherently Ill-posed problem. Vision Problem Revisited.

mali
Download Presentation

Stereo Vision using PatchMatch Algorithm

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. Stereo Vision using PatchMatch Algorithm Junkyung Kim Class of 2014

  2. 1. Introduction

  3. Vision Problem Revisited • Loss of information • 3-D physical world projected onto a 2-D surface

  4. Vision Problem Revisited • Therefore, an inherently Ill-posed problem

  5. Vision Problem Revisited • Possible solutions • 1. Directly exploit the Well-behavednessof the physical world • Shading • Occlusion • Textural transformation & alignment …

  6. Vision Problem Revisited • Possible solutions • 2. Rely on different sources of information • Ecolocation • “Kinect” …

  7. Vision Problem Revisited • Possible solutions • 2. Use multiple 2-D images • Motion • Stereo

  8. 1. Description of the problem

  9. What is Stereo Vision? • A constrained version of motion parallax • When the observer moves, closer objects appear to move faster • Constraint 1 : only two frames (binocular) • Constraint 2 : observer moves only laterally • Constraint 3 : all the objects are stationary

  10. What is Stereo Vision? • Binocular disparity for depth computation

  11. What is Stereo Vision? • How to get depth? = How to get delta-X ? • Must find the correspondence pairing first www.consortium.ri.cmu.edu

  12. What is Stereo Vision? • How to get depth? = How to get delta-X ?

  13. Correspondence Problem • Ambiguity, again • Exhaustive search, even though reduced only within the epipolar line, is highly prone to false pairing • 1. inherent imbalance : search space >> sample space • 2. noise

  14. Correspondence Problem • Ambiguity, again • Worst-case scenario : binary RDS

  15. Correspondence Problem • Solutions • 1. Use more evidence (neighboring pixels) • Equivalent to increasing sample space • Better-posed than ill-posed • 2. Enforce well-behavedness of the world • smoothness

  16. Correspondence Problem • Implementation • 1. Representational • Some feature map rather than raw image • 2. Computational • PatchMatch

  17. 3. Method : PatchMatch

  18. PatchMatch • A Randomized Correspondence Algorithm for Structural Image Editing • Barnes, et al • Patten Analysis & Recognition, 2009

  19. PatchMatch • Proposed Applications • Image reconstruction (reshuffling) Barnes et al, 2009

  20. PatchMatch • Proposed Applications • Image completion (inpainting) Barnes et al, 2009

  21. PatchMatch • Proposed Applications • Image retargeting (transformations) Barnes et al, 2009

  22. PatchMatch • Why is it suitable for stereo correspondence? • 1. patch-based match • Nearest-Neighbor Field (NNF) • large sample space > uniqueness constraint better satisfied

  23. PatchMatch • Why is it suitable for stereo correspondence? • 2. computational efficiency : • randomized search followed by propagation • Not just faster, but easier to implement smoothness enforcement • Important in application (e.g. navigation)

  24. 4. Milestones

  25. Milestones • Week 1~2 • Study PatchMatchand source code • Gather dataset to work on • Preferrably with one with ground-truth disparity map

  26. Milestones • Week 3~5 (mid-project presentation) • Preliminary implementation • Generate first-round of outputs

  27. Milestones • Week 6~9 (final presentation) • Evaluate the algorithm with full outputs • (if time allows) Draw comparison among several different implementations

More Related