相机标定(二)——图像坐标与世界坐标转换

相机标定(一)——内参标定与程序实现

相机标定(二)——图像坐标与世界坐标转换

相机标定(三)——手眼标定

一、坐标关系

相机标定(二)——图像坐标与世界坐标转换_第1张图片
相机中有四个坐标系,分别为worldcameraimagepixel

  • world为世界坐标系,可以任意指定 x w x_w xw轴和 y w y_w yw轴,为上图P点所在坐标系。
  • camera为相机坐标系,原点位于小孔,z轴与光轴重合, x w x_w xw轴和 y w y_w yw轴平行投影面,为上图坐标系 X c Y c Z c X_cY_cZ_c XcYcZc
  • image为图像坐标系,原点位于光轴和投影面的交点, x w x_w xw轴和 y w y_w yw轴平行投影面,为上图坐标系 X Y Z XYZ XYZ
  • pixel为像素坐标系,从小孔向投影面方向看,投影面的左上角为原点, u v uv uv轴和投影面两边重合,该坐标系与图像坐标系处在同一平面,但原点不同。

二、坐标变换

下式为像素坐标pixel与世界坐标world的变换公式,右侧第一个矩阵为相机内参数矩阵,第二个矩阵为相机外参数矩阵。
s [ u v 1 ] = [ f x 0 c x 0 f y c y 0 0 1 ] [ r 11 r 12 r 13 t 1 r 21 r 22 r 23 t 2 r 31 r 32 r 33 t 3 ] [ x y z 1 ] s\begin{bmatrix} u\\v\\1 \end{bmatrix} =\begin{bmatrix} f_x&0&c_x\\0&f_y&c_y\\0&0&1 \end{bmatrix} \begin{bmatrix} r_{11} & r_{12}&r_{13}&t_1 \\ r_{21} & r_{22}&r_{23}&t_2 \\ r_{31} & r_{32}&r_{33}&t_3 \end{bmatrix} \begin{bmatrix} x\\y\\z\\1 \end{bmatrix} suv1=fx000fy0cxcy1r11r21r31r12r22r32r13r23r33t1t2t3xyz1

2.1 变换流程

P u v = K T P w P_{uv} = KTP_w Puv=KTPw

该方程右侧隐含了一次齐次坐标到非齐次坐标的转换

  • K K K内参 T c a m e r a p i x e l T_{camera}^{pixel} Tcamerapixel:像素坐标系相对于相机坐标系的变换(与相机和镜头有关)
  • T T T外参 T w o r l d c a m e r a T_{world}^{camera} Tworldcamera:相机坐标系相对于世界坐标系的变换

顺序变换

  • pixelcamera,使用内参变换

P c a m e r a ( 3 × 1 ) = T c a m e r a p i x e l − 1 ( 3 × 3 ) ∗ P p i x e l ( 3 × 1 ) ∗ d e p t h P_{camera}(3 \times 1) = {T_{camera}^{pixel}}^{-1}(3 \times 3)*P_{pixel}(3 \times 1)*depth Pcamera(3×1)=Tcamerapixel1(3×3)Ppixel(3×1)depth

  • cameraworld,使用外参变换

P w o r l d ( 4 × 1 ) = T w o r l d c a m e r a − 1 ( 4 × 4 ) ∗ P c a m e r a ( 4 × 1 ) P_{world} (4 \times 1)= {T_{world}^{camera}}^{-1}(4 \times 4)* P_{camera}(4 \times 1) Pworld(4×1)=Tworldcamera1(4×4)Pcamera(4×1)

注意:两个变换之间的矩阵大小不同,需要分开计算,从pixelcamera获得的相机坐标为非齐次,需转换为齐次坐标再进行下一步变换。而在进行从cameraworld时,需将外参矩阵转换为齐次再进行计算。(齐次坐标的分析)

直接变换
[ X Y Z ] = R − 1 ( M − 1 ∗ s ∗ [ u v 1 ] − t ) \begin{bmatrix} X\\Y\\Z \end{bmatrix}= R^{-1}(M^{-1}*s*\begin{bmatrix} u\\v\\1 \end{bmatrix} - t) XYZ=R1(M1suv1t)
注意:直接变换是直接根据变换公式获得,实际上包含pixelcameracameraworld,实际上和顺序变换一样,通过顺序变换可以更清晰了解变换过程。

2.2 参数计算

  • 内参

通过张正友标定获得

  • 外参

通过PNP估计获得

  • 深度s

深度s为目标点在相机坐标系Z方向的值

2.3 外参计算

  • solvePnP函数

Perspective-n-Point是通过n组给定点的世界坐标与像素坐标估计相机位置的方法。OpenCV内部提供的函数为solvePnP(),函数介绍如下:

bool solvePnP(InputArray objectPoints, 
	      	  InputArray imagePoints, 
	      	  InputArray cameraMatrix, 
	      	  InputArray distCoeffs, 
              OutputArray rvec, 
              OutputArray tvec, 
              bool useExtrinsicGuess=false, 
              int flags=ITERATIVE )
  • objectPoints,输入世界坐标系中点的坐标;
  • imagePoints,输入对应图像坐标系中点的坐标;
  • cameraMatrix, 相机内参数矩阵;
  • distCoeffs, 畸变系数;
  • rvec, 旋转向量,需输入一个非空Mat,需要通过cv::Rodrigues转换为旋转矩阵;
  • tvec, 平移向量,需输入一个非空Mat;
  • useExtrinsicGuess, 默认为false,如果设置为true则输出输入的旋转矩阵和平移矩阵;
  • flags,选择采用的算法;
    • CV_ITERATIVE Iterative method is based on 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 projectPoints() ) objectPoints .
    • CV_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 four object and image points.
    • CV_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”.

注意solvePnP()的参数rvectvec应该都是double类型的

  • 程序实现
//输入参数
Mat cameraMatrix = Mat(3, 3, CV_32FC1, Scalar::all(0)); /* 摄像机内参数矩阵 */
Mat distCoeffs = Mat(1, 5, CV_32FC1, Scalar::all(0)); /* 摄像机的5个畸变系数:k1,k2,p1,p2,k3 */
double zConst = 0;//实际坐标系的距离,若工作平面与相机距离固定可设置为0
	
//计算参数
double s;
Mat rotationMatrix = Mat (3, 3, DataType::type);
Mat tvec = Mat (3, 1, DataType::type);
void calcParameters(vector imagePoints, vector objectPoints)
{
	//计算旋转和平移
	Mat rvec(3, 1, cv::DataType::type);
	cv::solvePnP(objectPoints, imagePoints, cameraMatrix, distCoeffs, rvec, tvec);
	cv::Rodrigues(rvec, rotationMatrix);
}

2.4 深度计算

理想情况下,相机与目标平面平行(只有绕Z轴的旋转),但实际上相机与目标平面不会完全平行,存在绕X和Y轴的旋转,此时深度s并不是固定值 t 3 t_3 t3,计算深度值为:
s = t 3 + r 31 ∗ x + r 32 ∗ y + r 33 ∗ z s = t_3 + r_{31} * x + r_{32} * y + r_{33} * z s=t3+r31x+r32y+r33z
若使用固定值进行变换会导致较大误差。解决方案如下:

  • 计算多个点的深度值,拟合一个最优值
  • 通过外参计算不同位置的深度(此处采用该方案)

注意:此处环境为固定单目与固定工作平面,不同情况下获得深度方法不同。

像素坐标pixel与世界坐标world转换公式可简化为
s [ u v 1 ] = M ( R [ X Y Z c o n s t ] + t ) s\begin{bmatrix} u\\v\\1 \end{bmatrix} =M(R\begin{bmatrix} X\\Y\\Z_{const} \end{bmatrix}+t) suv1=M(RXYZconst+t)
M M M为相机内参数矩阵, R R R为旋转矩阵, t t t为平移矩阵, z c o n s t z_{const} zconst为目标点在世界坐标Z方向的值,此处为0。

变换可得
R − 1 M − 1 s [ u v 1 ] = [ X Y Z c o n s t ] + R − 1 t R^{-1}M^{-1}s\begin{bmatrix} u\\v\\1 \end{bmatrix} =\begin{bmatrix} X\\Y\\Z_{const} \end{bmatrix}+R^{-1}t R1M1suv1=XYZconst+R1t
当相机内外参已知可计算获得 s s s

三、程序实现

3.1 Matlab

clc;
clear;

% 内参
syms fx cx fy cy;
M = [fx,0,cx;
    0,fy,cy;
    0,0,1];
            
% 外参
%旋转矩阵
syms r11 r12 r13 r21 r22 r23 r31 r32 r33;
R = [r11,r12,r13;
  	r21,r22,r23;
  	r31,r32,r33];

%平移矩阵
syms t1 t2 t3;
t = [t1;
    t2;
    t3];
%外参矩阵 
T = [R,t;
    0,0,0,1];
 
% 图像坐标 
syms u v;
imagePoint = [u;v;1];   
 
% 计算深度
syms zConst;
rightMatrix = inv(R)*inv(M)*imagePoint;
leftMatrix = inv(R)*t;
s = (zConst + leftTemp(3))/rightTemp(3);

% 转换世界坐标方式一
worldPoint1 = inv(R) * (s*inv(M) * imagePoint - t)

% 转换世界坐标方式二
cameraPoint = inv(M)* imagePoint * s;% image->camrea
worldPoint2 = inv(T)* [cameraPoint;1];% camrea->world
worldPoint2 = [worldPoint2(1);worldPoint2(2);worldPoint2(3)]

3.2 C++

该程序参考《视觉SLAM十四讲》第九讲实践章:设计前端代码部分进行修改获得,去掉了李群库Sopuhus依赖,因该库在windows上调用较为麻烦,若在Linux建议采用Sopuhus

  • camera.h
#ifndef CAMERA_H
#define CAMERA_H

#include 
#include 
using Eigen::Vector4d;
using Eigen::Vector2d;
using Eigen::Vector3d;
using Eigen::Quaterniond;
using Eigen::Matrix;

class Camera
{
public:
    Camera();

    // coordinate transform: world, camera, pixel
    Vector3d world2camera( const Vector3d& p_w);
    Vector3d camera2world( const Vector3d& p_c);
    Vector2d camera2pixel( const Vector3d& p_c);
    Vector3d pixel2camera( const Vector2d& p_p); 
    Vector3d pixel2world ( const Vector2d& p_p);
    Vector2d world2pixel ( const Vector3d& p_w);

	// set params
	void setInternalParams(double fx, double cx, double fy, double cy);
	void setExternalParams(Quaterniond Q, Vector3d t);
	void setExternalParams(Matrix  R, Vector3d t);

	// cal depth
	double calDepth(const Vector2d& p_p);

private:
    // 内参
	double fx_, fy_, cx_, cy_, depth_scale_;
	Matrix inMatrix_;

    // 外参
    Quaterniond Q_;
	Matrix  R_;
    Vector3d t_; 
	Matrix exMatrix_;
};

#endif // CAMERA_H
  • camera.cpp
#include "camera.h"

Camera::Camera(){}

Vector3d Camera::world2camera ( const Vector3d& p_w)
{
	Vector4d p_w_q{ p_w(0,0),p_w(1,0),p_w(2,0),1};
	Vector4d p_c_q = exMatrix_ * p_w_q;
	return Vector3d{p_c_q(0,0),p_c_q(1,0),p_c_q(2,0)};
}

Vector3d Camera::camera2world ( const Vector3d& p_c)
{
	Vector4d p_c_q{ p_c(0,0),p_c(1,0),p_c(2,0),1 };
	Vector4d p_w_q = exMatrix_.inverse() * p_c_q;
    return Vector3d{ p_w_q(0,0),p_w_q(1,0),p_w_q(2,0) };
}

Vector2d Camera::camera2pixel ( const Vector3d& p_c )
{
    return Vector2d (
               fx_ * p_c ( 0,0 ) / p_c ( 2,0 ) + cx_,
               fy_ * p_c ( 1,0 ) / p_c ( 2,0 ) + cy_
           );
}

Vector3d Camera::pixel2camera ( const Vector2d& p_p)
{
	double depth = calDepth(p_p);
    return Vector3d (
               ( p_p ( 0,0 )-cx_ ) *depth/fx_,
               ( p_p ( 1,0 )-cy_ ) *depth/fy_,
               depth
           );
}

Vector2d Camera::world2pixel ( const Vector3d& p_w)
{
    return camera2pixel ( world2camera(p_w) );
}

Vector3d Camera::pixel2world ( const Vector2d& p_p)
{
    return camera2world ( pixel2camera ( p_p ));
}

double Camera::calDepth(const Vector2d& p_p)
{
	Vector3d p_p_q{ p_p(0,0),p_p(1,0),1 };
	Vector3d rightMatrix = R_.inverse() * inMatrix_.inverse() * p_p_q;
	Vector3d leftMatrix = R_.inverse() * t_;
	return leftMatrix(2,0)/rightMatrix(2,0);
}

void Camera::setInternalParams(double fx, double cx, double fy, double cy)
{
	fx_ = fx;
	cx_ = cx;
	fy_ = fy;
	cy_ = cy;

	inMatrix_ << fx, 0, cx,
				0, fy, cy,
				0, 0, 1;
}

void Camera::setExternalParams(Quaterniond Q, Vector3d t)
{
	Q_ = Q;
	R_ = Q.normalized().toRotationMatrix();
	setExternalParams(R_,t);
}

void Camera::setExternalParams(Matrix  R, Vector3d t)
{
	t_ = t;
	R_ = R;

	exMatrix_ << R_(0, 0), R_(0, 1), R_(0, 2), t(0,0),
		R_(1, 0), R_(1, 1), R_(1, 2), t(1,0),
		R_(2, 0), R_(2, 1), R_(2, 2), t(2,0),
		0, 0, 0, 1;
}

参考

image coordinate to world coordinate opencv

Computing x,y coordinate (3D) from image point

单应矩阵

camera_calibration_and_3d

《视觉SLAM十四讲》—相机与图像+实践章:设计前端

你可能感兴趣的:(视觉学习)