Download presentation
Presentation is loading. Please wait.
1
Evaluation of features detectors and descriptors based on 3D objects P. Moreels - P. Perona California Institute of Technology
2
Large baseline stereo Features – what for ? Stitching Object recognition [Dorko & Schmid’05] [Lowe ’04] [Brown & Lowe ’03] [Tuytelaars & Van Gool ’00]
3
Moving the viewpoint
4
232 keypoints extracted Features stability Features stability is not perfect… 240 keypoints extracted
5
First stage – feature detector difference of gaussians [Crowley’84] Kadir & Brady [Kadir’02] Harris [Harris’88] Affine invariant Harris [Mikolajczyk’02]
6
Second stage – feature descriptor SIFTSteerable filters Differential invariants Shape context [Lowe ’04][Freeman’91 ] [Belongie’02 ][Schmid’97 ]
7
Evaluations – Mikolajczyk ’03-’05 Large viewpoint change Computation of ground truth positions via a homography
8
Evaluations – Mikolajczyk ’03-’05 [CVPR’03][PAMI’04] [submitted] SIFT-based descriptors rule ! All affine-invariant detectors are good, they should all be used together.
9
2D vs. 3D Ranking of detectors/descriptors combinations are modified when switching from 2D to 3D objects
10
Dataset – 100 3D objects
11
Viewpoints 45° apart
15
Ground truth - Epipolar constraints
16
Testing setup Unrelated images used to load the database of features.
17
Distance ratio Correct matches are highly distinctive lower ratio Incorrect correspondences are ‘random correspondences’ low distinctiveness and ratio close to 1 [Lowe’04]
18
Are we accepting wrong matches ? Manual user classification into correct and incorrect triplets Comparison with a simpler system: 2 views, only one epipolar constraint. Pietro said maybe don’t need this slide – I think it is important to justify our 3-cameras setup
19
Detectors / descriptors tested DetectorsDescriptors Harris Hessian Harris-affine Hessian-affine Difference-of-gaussians MSER Kadir-Brady SIFT steerable filters differential invariants shape context PCA-SIFT
20
Results – viewpoint change Mahalanobis distance No ‘background’ images
21
Results – lighting / scale changes Change in light – result averaged over 3 lighting conditions. Change in scale - 7.0mm to 14.6mm
22
Conclusions Automated ground truth for 3D objects/scenes Ranking changes from 2D to 3D Stability is much lower for 3D Detectors – affine-rectified detectors are indeed best Descriptors – SIFT and shape context performed best. Application: use ground truth in order to learn probability densities: ‘how does a correct match look like ?’
Similar presentations
© 2025 SlidePlayer.com. Inc.
All rights reserved.