【论文阅读】EPnP: An Accurate O(n) Solution to the PnP Problem

EPnP: An Accurate O(n) Solution to the PnP Problem

Opencv.solvePnP documentation

solvePnP

bool cv::solvePnP(
InputArray 	objectPoints,
InputArray 	imagePoints,
InputArray 	cameraMatrix,
InputArray 	distCoeffs,
OutputArray 	rvec,
OutputArray 	tvec,
bool 	useExtrinsicGuess = false,
int 	flags = SOLVEPNP_ITERATIVE 
)

Finds an object pose from 3D-2D point correspondences.
This function returns the rotation and the translation vectors that transform a 3D point expressed in the object coordinate frame to the camera coordinate frame, using different methods:

• P3P methods (SOLVEPNP_P3P, SOLVEPNP_AP3P): need 4 input points to return a unique solution.
• SOLVEPNP_IPPE Input points must be >= 4 and object points must be coplanar(共面).
• SOLVEPNP_IPPE_SQUARE Special case suitable for marker pose estimation. Number of input points must be 4. Object points must be defined in the following order:
  point 0: [-squareLength / 2, squareLength / 2, 0]
  point 1: [ squareLength / 2, squareLength / 2, 0]
  point 2: [ squareLength / 2, -squareLength / 2, 0]
  point 3: [-squareLength / 2, -squareLength / 2, 0]
• for all the other flags, number of input points must be >= 4 and object points can be in any configuration.

Flags of solvePnP

• cv::SOLVEPNP_ITERATIVE Default solution. Iterative method is based on a Levenberg-Marquardt optimization. In this case the function finds such a pose that minimizes reprojection error, that is the sum of squared distances between the observed projections “imagePoints” and the projected (using cv::projectPoints ) “objectPoints”. Initial solution for non-planar(非平面) “objectPoints” needs at least 6 points and uses the DLT algorithm. Initial solution for planar “objectPoints” needs at least 4 points and uses pose from homography decomposition（单应性分解）.

• cv::SOLVEPNP_P3P Method is based on the paper of X.S. Gao, X.-R. Hou, J. Tang, H.-F. Chang “Complete Solution Classification for the Perspective-Three-Point Problem”. In this case the function requires exactly 4 object and image points.=>same as use solveP3P with same flag

• cv::SOLVEPNP_AP3P Method is based on the paper of T. Ke, S. Roumeliotis “An Efficient Algebraic Solution to the Perspective-Three-Point Problem” . In this case the function requires exactly 4 object and image points.=>same as use solveP3P with same flag

• cv::SOLVEPNP_EPNP Method has been introduced by F. Moreno-Noguer, V. Lepetit and P. Fua in the paper “EPnP: Efficient Perspective-n-Point Camera Pose Estimation” .

• cv::SOLVEPNP_DLS Broken implementation. Using this flag will fallback to EPnP.

• cv::SOLVEPNP_UPNP Broken implementation. Using this flag will fallback to EPnP.

• cv::SOLVEPNP_IPPE Method is based on the paper of T. Collins and A. Bartoli. “Infinitesimal Plane-Based Pose Estimation”. This method requires coplanar object points.

• cv::SOLVEPNP_IPPE_SQUARE Method is based on the paper of Toby Collins and Adrien Bartoli. “Infinitesimal Plane-Based Pose Estimation”. This method is suitable for marker pose estimation. It requires 4 coplanar object points defined in the following order:
  point 0: [-squareLength / 2, squareLength / 2, 0]
  point 1: [ squareLength / 2, squareLength / 2, 0]
  point 2: [ squareLength / 2, -squareLength / 2, 0]
  point 3: [-squareLength / 2, -squareLength / 2, 0]

• cv::SOLVEPNP_SQPNP Method is based on the paper “A Consistently Fast and Globally Optimal Solution to the Perspective-n-Point Problem” by G. Terzakis and M.Lourakis. It requires 3 or more points.

similar functions

1. cv::solveP3P()
   Finds an object pose from 3 3D-2D point correspondences.

2. cv::solvePnPGeneric()

Finds an object pose from 3D-2D point correspondences.

flags:

• P3P methods (SOLVEPNP_P3P, SOLVEPNP_AP3P)
• SOLVEPNP_IPPE
• SOLVEPNP_IPPE_SQUARE

3. solvePnPRansac()
   using the RANSAC scheme

4. solvePnPRefineLM()
   Refine a pose (the translation and the rotation that transform a 3D point expressed in the object coordinate frame to the camera coordinate frame) from a 3D-2D point correspondences and starting from an initial solution.

5. solvePnPRefineVVS()
   Refine a pose (the translation and the rotation that transform a 3D point expressed in the object coordinate frame to the camera coordinate frame) from a 3D-2D point correspondences and starting from an initial solution.

cv::SOLVEPNP_EPNP: Paper 008

论文内容：

propose a non-iterative solution to the PnP problem—the estimation of the pose of a cali- brated camera from n 3D-to-2D point correspondences— whose computational complexity grows linearly with n.

EPNP是求解PnP问题的非迭代方法。
优势是:

1. 时间复杂度低，O（n），其他的是指数级，所以耗时短；
2. 不依赖初始化；
3. 鲁棒性高；
4. 误差小。
5. 用高斯牛顿方法可以进一步提高准确率，但计算耗时不会太大。

看论文里的实验数据，EPNP方法求得的rotation和translation存在随着匹配点数的增多，逐渐收敛趋于稳定的过程

图上看的话，噪声恒定的情况下，差不多10个点左右趋于收敛。