1 / 39

EE 7700

EE 7700. Geometric Transformations. Geometric Transformation. translation. Rotation matrix. scale. Scale matrix. rotation & scale. Rigid flow. Affine Flow. Perspective Flow. EE 7730. RANSAC: RANdom SAmple Consensus. Outliers. Consider the least squares fit for optical flow:.

Télécharger la présentation

EE 7700

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. EE 7700 Geometric Transformations

  2. Geometric Transformation translation Rotation matrix scale Scale matrix rotation & scale Rigid flow

  3. Affine Flow

  4. Perspective Flow

  5. EE 7730 RANSAC: RANdom SAmple Consensus

  6. Outliers • Consider the least squares fit for optical flow: If some of the values are wrong, it will degrade the estimation.

  7. Outliers • It is best not to include outliers in the estimation. Line Fitting Problem: Given (x1,y1), …, (xN,yN); find the line y=ax+b Outliers Best fit is degraded due to the outliers

  8. Robust Estimation • Two step process: • Classify data points as outliers or inliers • Use inliers only to fit a model

  9. RANSAC • Repeat for k times: • Randomly choose n points (the smallest number of points required) from the data. • Estimate the parameters using these points. • For each data point other than these n points: • Check if the data point is within a threshold, t, distance of current model; if it is, the data point is consistent with current model. • The total number of data points that are consistent is model’s support. • If the support is larger than a predetermined number, d, then there is a good fit. Re-estimate the parameters using these inliers. • Choose the best fit with the smallest fitting error.

  10. RANSAC Two samples and their supports for line-fitting

  11. Example • Find the perspective parameters from Hartley & Zisserman

  12. Example • Apply corner detectors to both images from Hartley & Zisserman

  13. Example • Find the best match within a search window. from Hartley & Zisserman

  14. Example • Initial match results from Hartley & Zisserman 188 matched features in left image pointing to locations of corresponding right image features

  15. Example • Inliers and outliers after RANSAC from Hartley & Zisserman 89 outliers 99 inliers

  16. Panoramic Image Reconstruction Find features Match features Fit parametric model Application: Mosaic construction

  17. EE7730 Stereo Vision

  18. Pinhole Camera

  19. Review: Perspective Projection

  20. Stereo scene point p p’ image plane optical center p p’

  21. Epipolar Line p’ Y2 X2 Z2 O2 Epipole Stereo Constraints M Image plane Y1 p O1 Z1 X1 Focal plane

  22. P p p’ O’ O From Geometry to Algebra All vectors shown lie on the same plane.

  23. P p p’ O’ O From Geometry to Algebra

  24. Matrix form of cross product a=axi+ayj+azk a×b=|a||b|sin(η)u b=bxi+byj+bzk

  25. The Essential Matrix Essential matrix

  26. disparity Depth Z Elevation Zw A Simple Stereo System LEFT CAMERA RIGHT CAMERA baseline Right image: target Left image: reference Zw=0

  27. Parallel Cameras P Z xl xr f pl pr Ol Or Disparity: T T is the stereo baseline

  28. Stereo View Right View Left View Disparity

  29. (xl, yl) Correlation Approach LEFT IMAGE • For Each point (xl, yl) in the left image, define a window centered at the point

  30. Correlation Approach RIGHT IMAGE (xl, yl) • … search its corresponding point within a search region in the right image

  31. Correlation Approach RIGHT IMAGE (xr, yr) dx (xl, yl) • … the disparity (dx, dy) is the displacement when the correlation is maximum

  32. Stereo results • Data from University of Tsukuba Scene Ground truth (Seitz)

  33. Results with window correlation Estimated depth of field Ground truth (Seitz)

  34. Results with better method • A state of the art method • Boykov et al., Fast Approximate Energy Minimization via Graph Cuts, • International Conference on Computer Vision, September 1999. Ground truth (Seitz)

  35. Applications First-down line courtesy of Sportvision

  36. Applications Virtual advertising courtesy of Princeton Video Image

More Related