1 Comparative Survey on Fundamental Matrix Estimation Computer Vision and Robotics Group Institute of Informatics and Applications University of Girona.

Slides:



Advertisements
Similar presentations
Zhengyou Zhang Vision Technology Group Microsoft Research
Advertisements

The fundamental matrix F
Lecture 11: Two-view geometry
Institut für Elektrische Meßtechnik und Meßsignalverarbeitung Professor Horst Cerjak, Augmented Reality VU 2 Calibration Axel Pinz.
MASKS © 2004 Invitation to 3D vision Lecture 7 Step-by-Step Model Buidling.
Robot Vision SS 2005 Matthias Rüther 1 ROBOT VISION Lesson 3: Projective Geometry Matthias Rüther Slides courtesy of Marc Pollefeys Department of Computer.
Chapter 6 Feature-based alignment Advanced Computer Vision.
Two-view geometry.
Camera calibration and epipolar geometry
Structure from motion.
Multiple View Geometry
Robot Vision SS 2008 Matthias Rüther 1 ROBOT VISION Lesson 6: Shape from Stereo Matthias Rüther Slides partial courtesy of Marc Pollefeys Department of.
Computing F Class 8 Read notes Section 4.2. C1C1 C2C2 l2l2  l1l1 e1e1 e2e2 Fundamental matrix (3x3 rank 2 matrix) 1.Computable from corresponding points.
Epipolar geometry. (i)Correspondence geometry: Given an image point x in the first view, how does this constrain the position of the corresponding point.
Structure from motion. Multiple-view geometry questions Scene geometry (structure): Given 2D point matches in two or more images, where are the corresponding.
Uncalibrated Geometry & Stratification Sastry and Yang
Lecture 21: Multiple-view geometry and structure from motion
Multiple View Geometry Marc Pollefeys University of North Carolina at Chapel Hill Modified by Philippos Mordohai.
Multiple View Geometry
Many slides and illustrations from J. Ponce
Geometric Optimization Problems in Computer Vision.
Lecture 11: Structure from motion CS6670: Computer Vision Noah Snavely.
Multiple View Geometry Marc Pollefeys University of North Carolina at Chapel Hill Modified by Philippos Mordohai.
Independent Motion Estimation Luv Kohli COMP Multiple View Geometry May 7, 2003.
Single-view geometry Odilon Redon, Cyclops, 1914.
Projected image of a cube. Classical Calibration.
May 2004Stereo1 Introduction to Computer Vision CS / ECE 181B Tuesday, May 11, 2004  Multiple view geometry and stereo  Handout #6 available (check with.
1Jana Kosecka, CS 223b EM and RANSAC EM and RANSAC.
Multiple View Geometry
Global Alignment and Structure from Motion Computer Vision CSE455, Winter 2008 Noah Snavely.
CSCE 641 Computer Graphics: Image-based Modeling (Cont.) Jinxiang Chai.
Multiple View Geometry. THE GEOMETRY OF MULTIPLE VIEWS Reading: Chapter 10. Epipolar Geometry The Essential Matrix The Fundamental Matrix The Trifocal.
55:148 Digital Image Processing Chapter 11 3D Vision, Geometry Topics: Basics of projective geometry Points and hyperplanes in projective space Homography.
CSE 6367 Computer Vision Stereo Reconstruction Camera Coordinate Transformations “Everything should be made as simple as possible, but not simpler.” Albert.
Multi-view geometry. Multi-view geometry problems Structure: Given projections of the same 3D point in two or more images, compute the 3D coordinates.
776 Computer Vision Jan-Michael Frahm, Enrique Dunn Spring 2013.
Automatic Camera Calibration
Chapter 6 Feature-based alignment Advanced Computer Vision.
Introduction à la vision artificielle III Jean Ponce
Computing the Fundamental matrix Peter Praženica FMFI UK May 5, 2008.
Multi-view geometry.
1 Robust estimation techniques in real-time robot vision Ezio Malis, Eric Marchand INRIA Sophia, projet ICARE INRIA Rennes, projet Lagadic.
Projective cameras Motivation Elements of Projective Geometry Projective structure from motion Planches : –
CSCE 643 Computer Vision: Structure from Motion
3D Reconstruction Jeff Boody. Goals ● Reconstruct 3D models from a sequence of at least two images ● No prior knowledge of the camera or scene ● Use the.
Multiview Geometry and Stereopsis. Inputs: two images of a scene (taken from 2 viewpoints). Output: Depth map. Inputs: multiple images of a scene. Output:
Robot Vision SS 2007 Matthias Rüther 1 ROBOT VISION Lesson 6a: Shape from Stereo, short summary Matthias Rüther Slides partial courtesy of Marc Pollefeys.
© 2005 Martin Bujňák, Martin Bujňák Supervisor : RNDr.
Computing F. Content Background: Projective geometry (2D, 3D), Parameter estimation, Algorithm evaluation. Single View: Camera model, Calibration, Single.
Geometry of Multiple Views
Ch. 3: Geometric Camera Calibration
A Flexible New Technique for Camera Calibration Zhengyou Zhang Sung Huh CSPS 643 Individual Presentation 1 February 25,
EECS 274 Computer Vision Geometric Camera Calibration.
Two-view geometry. Epipolar Plane – plane containing baseline (1D family) Epipoles = intersections of baseline with image planes = projections of the.
Feature Matching. Feature Space Outlier Rejection.
55:148 Digital Image Processing Chapter 11 3D Vision, Geometry Topics: Basics of projective geometry Points and hyperplanes in projective space Homography.
3D Reconstruction Using Image Sequence
Robust Estimation Course web page: vision.cis.udel.edu/~cv April 23, 2003  Lecture 25.
Lec 26: Fundamental Matrix CS4670 / 5670: Computer Vision Kavita Bala.
Correspondence and Stereopsis. Introduction Disparity – Informally: difference between two pictures – Allows us to gain a strong sense of depth Stereopsis.
Multi-view geometry. Multi-view geometry problems Structure: Given projections of the same 3D point in two or more images, compute the 3D coordinates.
Geometric Camera Calibration
55:148 Digital Image Processing Chapter 11 3D Vision, Geometry
Two-view geometry Computer Vision Spring 2018, Lecture 10
Epipolar geometry.
Estimating 2-view relationships
Two-view geometry.
Two-view geometry.
Multi-view geometry.
Parameter estimation class 6
Presentation transcript:

1 Comparative Survey on Fundamental Matrix Estimation Computer Vision and Robotics Group Institute of Informatics and Applications University of Girona. Girona (Spain) {armangue, jpages, qsalvi, X. Armangué, J. Pagès, J. Salvi and J. Batlle

2 Contents : 1.- Stereo Vision 2.- The Epipolar Geometry 3.- Computing the Fundamental Matrix Linear methods Iterative methods Robust methods 4.- Experimental Results

3 The principle of Triangulation 3D point Stereo Vision

4 T21 = [R1(:,1:3) T1]; invT21=inv(T21); P2Dw1=invT21*[Xu1; Yu1; f1; 1]; Ocw1=invT21(:,4); T22 = [R2(:,1:3) T2]; invT22=inv(T22); P2Dw2=invT22*[Xu2; Yu2; f2; 1] Ocw2=invT22(:,4); W Oc 2 {W} W Oc 1 W P2D 2 W P2D 1 u v pq pq=Ocw2(1:3)-Ocw1(1:3); u=P2Dw1(1:3,i)-Ocw1(1:3); v=P2Dw2(1:3,i)-Ocw2(1:3); alpha=(pq'*v-(pq'*u)*norm(v)^2/(u'*v))/((u'*v)-norm(u)^2*norm(v)^2/(u'*v)); beta=(-pq'*u+alpha*norm(u)^2)/(u'*v); r=Ocw1(1:3)+alpha.*u; s=Ocw2(1:3)+beta.*v; P3Dstereo = (r+s)./2; disterror = norm(r-s); W P3D Getting the 3D point Stereo Vision

5 Camera Pose Stereo Vision 3D Reconstruction: Optics and Internal Geometry Constraints: The Correspondence Problem. Active Systems: Non static Camera Position nor Orientation Epipolar Geometry I’I’ I OC’OC’ OCOC m m’m’ M OIOI OI’OI’ OWOW

6 The Epipolar Geometry O W coincides with O C ’ I’I’ I OC’OC’ OCOC m m’m’ e e’e’ M OIOI OI’OI’ lm’lm’ l’ml’m  OWOW CKC’CKC’ Intrinsic Extrinsic

7 Epipole Epipolar lines Area 1 Area 2 Correspondence points Zoom Area 1 Zoom Area 2 Epipolar geometry of Camera 1 Epipolar geometry of Camera 2 The Epipolar Geometry

8 Computing the Fundamental Matrix: The Survey Linear Methods. Iterative Methods. Robust Methods. DrivingINRIAAerialUnderwater

9 LinearIterativeRobustOptimisationRank-2 Seven points (7p) X—yes Eight points (8p) XLS or Eig.no Rank-2 constraint XLSyes Iterative Newton- Raphson XLSno Linear iterative XLSno Non-linear minimization in parameter space XEig.yes Gradient technique XLS or Eig.no M-Estimator XLS or Eig.no / yes LMedS X7p / LS or Eig.no RANSAC X7p / Eigno LS: Least-Squares Eig: Eigen Analysis Computing the Fundamental Matrix: Survey

10 Computing the Fundamental Matrix: Linear Methods Seven points  It depends extremely on the seven points used. Eight points Least-squares minimization Eigen analysis  Better results increasing the number of points.  Eigen minimization is more realistic. Analytical method with Rank-2 constraint.  Forces a unique epipole, but results do not improve. Least-squares Eigen Analysis

11 Linear Methods Seven Points 8-points Least Squares 8-points Eigen Analysis Computing Noise Outliers X X

12 Computing the Fundamental Matrix: Iterative Methods Iterative Newton-Raphson  Good results. Depends on the initial guess. Linear iterative method.  i is based on the F of the previous step. F is computed by using Least-squares in each iteration.  Improves linear least-squares considerably. Non-linear minimization in parameter space  Forces a Rank-2 F but the discrepancy is high. Gradient technique Least-squares or Eigen Analysis.  Better results with eigen analysis.

13 Iterative Methods Newton-Raphson Linear minimization Forcing Rank-2 Computing Noise Outliers X X

14 Computing the Fundamental Matrix: Robust Methods Robust Methods: M-Estimator  Reduces effect of outliers weighting the residual of each point.  Lots of methods just defining a new weight-function. Leasts-squares, eigen analysis, Torr, etc...  Good results in the presence gaussian noise in point localization.  Bad results in the presence of outliers. LMedS & RANSAC  Points used to compute F randomly selected.  LMedS uses the median of distances.  RANSAC maximises de number of inliers.  LMedS is more restrictive than RANSAC (removes more points).  Once the outliers are removed, F is recalculated.

15 Robust Methods Initial Matching

16 Robust Methods: M-estimator Torr

17 Robust Methods: RANSAC

18 Robust Methods: LMedS Eigen Computing Noise Outliers X X

19 Methods Implemented with mean and std. of error: 1.- seven points; 2.- least-squares (LS); 3.- orthogonal LS; 4.- rank-2 constraint; 5.- iterative lineal using LS; 6.- iterative Newton-Raphson using LS; 7.- minimization in parameter space using eigen; 8.- gradient using LS; 9.- gradient using eigen; 10.- M-Estimator using LS; 11.- M-Estimator using eigen; 12.- M-Estimator proposed by Torr; 13.- LMedS using LS; 14.- LMedS using eigen; 15.- RANSAC using eigen. LinearIterativeRobust Computing the Fundamental Matrix: Results

20 Computing the Fundamental Matrix: Time Linear Iterative Robust

21 Conclusions Survey of fifteen methods of computing F. Conditions: Gaussian Noise, Outliers and Real Images. Linear methods: Good results if the points are well located and the correspondence problem previously solved (without outliers). Iterative methods: Can cope with noise but inefficient in the presence of outliers. Robust methods: Cope with both noise and outliers. Eigen Analysis is better than least-squares. Rank-2 matrices are preferred if a good geometry is required. Better results if data is previously normalized. traslated so that they centroid is placed at the origin. Scaled so that the mean of distances to the origin is [-1, 1] Code available: