EPnP: An Accurate O(n) Solution to the PnP Problem
read more
Citations
ORB: An efficient alternative to SIFT or SURF
ORB-SLAM: A Versatile and Accurate Monocular SLAM System
ORB-SLAM: a Versatile and Accurate Monocular SLAM System
Structure-from-Motion Revisited
VINS-Mono: A Robust and Versatile Monocular Visual-Inertial State Estimator
References
Random sample consensus: a paradigm for model fitting with applications to image analysis and automated cartography
Multiple view geometry in computer vision
Multiple View Geometry in Computer Vision.
Least-Squares Fitting of Two 3-D Point Sets
Related Papers (5)
Frequently Asked Questions (14)
Q2. What contributions have the authors mentioned in the paper "Epnp: accurate non-iterative o(n) solution to the pnp problem" ?
The authors propose a non-iterative solution to the PnP problem—the estimation of the pose of a calibrated camera from n 3D-to-2D point correspondences—whose computational complexity grows linearly with n. Furthermore, if maximal precision is required, the output of the closed-form solution can be used to initialize a Gauss-Newton scheme, which improves accuracy with negligible amount of additional time.
Q3. How do the authors refine the coefficients in Eq. 8?
The authors refine the four values β = [β1, β2, β3, β4] of the coefficients in Eq. 8 by choosing the values that minimize the change in distance between control points.
Q4. How do the authors estimate the unknown depths of the polynomials?
The coefficients of these polynomials are then arranged in a (n−1)(n−2) 2× 5 matrix and singular value decomposition (SVD) is used to estimate the unknown depths.
Q5. What are the unknown parameters of this linear system?
(4)The unknown parameters of this linear system are the 12 control point coordinates { (xcj, y c j , z c j) } j=1,...,4and the n projective parameters {wi}i=1,...,n.
Q6. What is the way to exploit the intrinsic parameters?
A way to exploit their knowledge of the intrinsic parameters is to clamp the retrieved values to the known ones, but the accuracy still remains low.
Q7. How can the authors find the correct value of the ii=1?
Given that the solution can be expressed as a linear combination of the null eigenvectors of M M, finding it amounts to computing the appropriate values for the {βi}i=1,...,N coefficients of Eq. 8.
Q8. How fast is the ambiguity in the camera pose?
For instance, for n = 6 points, their algorithm is about 10 times faster than LHM, and about 200 times faster than AD.2) The Planar Case: Schweighofer and Pinz [25] prove that when the reference points lie on a plane, camera pose suffers from an ambiguity that results in significant instability.
Q9. What is the weighted sum of the null eigenvectors of M?
More precisely, it is a weighted sum of the null eigenvectors of M. Given that the correct linear combination is the one that yields 3D camera coordinates for the control points that preserve their distances, the authors can find the appropriate weights by solving small systems of quadratic equations, which can be done at a negligible computational cost.
Q10. How many smallest eigenvalues do the camera have?
as the focal length increases and the camera becomes closer to being orthographic, all four smallest eigenvalues approach zero.
Q11. How many eigenvalues can be used to compute the distances between control points?
4. In this section, the authors show that the fact that the distances between control points must be preserved can be expressed in terms of a small number of quadratic equations, which can be efficiently solved to compute {βi}i=1,...,N for N = 1, 2, 3 and 4.
Q12. How many points are needed to estimate the pose?
Even if four correspondences are sufficient in general to estimate the pose, it is nonetheless desirable to consider larger point sets to introduce redundancy and reduce the sensitivity to noise.
Q13. How do the authors find the stability of their method?
in practice, the authors have found that the stability of their method is increased by taking the centroid of the reference points as one, and to select the rest in such a way that they form a basis aligned with the principal directions of the data.
Q14. Why did Schweighofer and Pinz omit the EPnP+GN?
The authors omit as well the EPnP+GN, because for the planar case the closed-form solution for the non-ambiguous cases was already very accurate, and the Gauss-Newton optimization could not help to resolve the ambiguity in the rest of cases.