1 / 21

Nearest-neighbor matching to feature database

Nearest-neighbor matching to feature database. Hypotheses are generated by matching each feature to nearest neighbor vectors in database No fast method exists for always finding 128-element vector to nearest neighbor in a large database Therefore, use approximate nearest neighbor:

kbodily
Download Presentation

Nearest-neighbor matching to feature database

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. Nearest-neighbor matching to feature database • Hypotheses are generated by matching each feature to nearest neighbor vectors in database • No fast method exists for always finding 128-element vector to nearest neighbor in a large database • Therefore, use approximate nearest neighbor: • We use best-bin-first (Beis & Lowe, 97) modification to k-d tree algorithm • Use heap data structure to identify bins in order by their distance from query point • Result: Can give speedup by factor of 1000 while finding nearest neighbor (of interest) 95% of the time

  2. Detecting 0.1% inliers among 99.9% outliers • Need to recognize clusters of just 3 consistent features among 3000 feature match hypotheses • LMS or RANSAC would be hopeless! • Use generalized Hough transform • Vote for each potential match according to model ID and pose • Insert into multiple bins to allow for error in similarity approximation • Using a hash table instead of an array avoids need to form empty bins or predict array size

  3. Probability of correct match • Compare distance of nearest neighbor to second nearest neighbor (from different object) • Threshold of 0.8 provides excellent separation

  4. Model verification • Examine all clusters in Hough transform with at least 3 features • Perform least-squares affine fit to model. • Discard outliers and perform top-down check for additional features. • Evaluate probability that match is correct • Use Bayesian model, with probability that features would arise by chance if object was not present • Takes account of object size in image, textured regions, model feature count in database, accuracy of fit (Lowe, CVPR 01)

  5. Solution for affine parameters • Affine transform of [x,y] to [u,v]: • Rewrite to solve for transform parameters:

  6. Models for planar surfaces with SIFT keys Planar texture models

  7. Planar recognition • Planar surfaces can be reliably recognized at a rotation of 60° away from the camera • Affine fit approximates perspective projection • Only 3 points are needed for recognition

  8. 3D Object Recognition • Extract outlines with background subtraction

  9. 3D Object Recognition • Only 3 keys are needed for recognition, so extra keys provide robustness • Affine model is no longer as accurate

  10. Recognition under occlusion

  11. Test of illumination invariance • Same image under differing illumination 273 keys verified in final match

  12. Examples of view interpolation

  13. Recognition using View Interpolation

  14. Location recognition

  15. Robot Localization • Joint work with Stephen Se, Jim Little

  16. Map continuously built over time

  17. Locations of map features in 3D

  18. Recognizing Panoramas • Matthew Brown and David Lowe • Recognize overlap from an unordered set of images and automatically stitch together • SIFT features provide initial feature matching • Image blending at multiple scales hides the seams Panorama of our lab automatically assembled from 143 images

  19. Multiple panoramas from an unordered image set

  20. Image registration and blending

  21. Comparison to template matching • Costs of template matching • 250,000 locations x 30 orientations x 4 scales = 30,000,000 evaluations • Does not easily handle partial occlusion and other variation without large increase in template numbers • Viola & Jones cascade must start again for each qualitatively different template • Costs of local feature approach • 3000 evaluations (reduction by factor of 10,000) • Features are more invariant to illumination, 3D rotation, and object variation • Use of many small subtemplates increases robustness to partial occlusion and other variations

More Related