图优化理论简介
我们已经介绍了非线性最小二乘的求解方式。它们是由很多个误差项之和组成的。然而,仅有一组优化变量和许多个误差项,我们并不清楚它们之间的关联。比方说,某一个优化变量 x j 存在于多少个误差项里呢?我们能保证对它的优化是有意义的吗?,我们希望能够直观地看到该优化问题长什么样。于是,就说到了图优化。图优化,是把优化问题表现成图(Graph)的一种方式。这里的图是图论意义上的图。一个图由若干个顶点(Vertex),以及连接着这些节点的边(Edge)组成。进而,用顶点表示优化变量,用边表示误差项。
曲线拟合示例1
代码
#include <iostream>
#include <g2o/core/base_vertex.h>
#include <g2o/core/base_unary_edge.h>
#include <g2o/core/block_solver.h>
#include <g2o/core/optimization_algorithm_levenberg.h>
#include <g2o/core/optimization_algorithm_gauss_newton.h>
#include <g2o/core/optimization_algorithm_dogleg.h>
#include <g2o/solvers/dense/linear_solver_dense.h>
#include <Eigen/Core>
#include <opencv2/core/core.hpp>
#include <cmath>
#include <chrono>
using namespace std;
// 曲线模型的顶点,模板参数:优化变量维度和数据类型--Vertex顶点
//(//第一个参数3表示update指针指向的数组大小,Eigen::Vector3d为_estimate的类型)
class CurveFittingVertex: public g2o::BaseVertex<3, Eigen::Vector3d> //基顶点 BaseVertex
{
public:
EIGEN_MAKE_ALIGNED_OPERATOR_NEW
virtual void setToOriginImpl() // 重置,顶点的重置函数
{
_estimate << 0,0,0;
}
virtual void oplusImpl( const double* update ) // 更新,顶点的更新函数
{
_estimate += Eigen::Vector3d(update);
}
// 存盘和读盘:留空
virtual bool read( istream& in ) {}
virtual bool write( ostream& out ) const {}
};
// 误差模型 模板参数:观测值维度,类型,连接顶点类型--Edge边
//(//1为_error的大小,double为观测值_measurement的大小,CurveFittingVetex为一元边上的一个顶点)
class CurveFittingEdge: public g2o::BaseUnaryEdge<1,double,CurveFittingVertex>
{
public:
EIGEN_MAKE_ALIGNED_OPERATOR_NEW
CurveFittingEdge( double x ): BaseUnaryEdge(), _x(x) {}
// 计算曲线模型误差,边的误差计算函数
void computeError()
{
const CurveFittingVertex* v = static_cast<const CurveFittingVertex*> (_vertices[0]);
const Eigen::Vector3d abc = v->estimate();
_error(0,0) = _measurement - std::exp( abc(0,0)*_x*_x + abc(1,0)*_x + abc(2,0) ) ;
}
virtual bool read( istream& in ) {}
virtual bool write( ostream& out ) const {}
public:
double _x; // x 值, y 值为 _measurement
};
int main( int argc, char** argv )
{
double a=1.0, b=2.0, c=1.0; // 真实参数值
int N=100; // 数据点
double w_sigma=1.0; // 噪声Sigma值
cv::RNG rng; // OpenCV随机数产生器
double abc[3] = {0,0,0}; // abc参数的估计值
vector<double> x_data, y_data; // 数据
cout<<"generating data: "<<endl;
for ( int i=0; i<N; i++ )
{
double x = i/100.0;
x_data.push_back ( x );
y_data.push_back (
exp ( a*x*x + b*x + c ) + rng.gaussian ( w_sigma )
);
cout<<x_data[i]<<" "<<y_data[i]<<endl;
}
//步骤就是,先把复杂的矩阵块类型在这里定义为Block.然后定义线性方程求解器linearsolver,再定义矩阵块求解器solver_ptr和linearsolver有关,再定义梯度下降方法solver和solver_ptr有关。
// 构建图优化,先设定g2o
typedef g2o::BlockSolver< g2o::BlockSolverTraits<3,1> > Block; // 每个误差项优化变量维度为3,误差值维度为1
Block::LinearSolverType* linearSolver = new g2o::LinearSolverDense<Block::PoseMatrixType>(); // 线性方程求解器
Block* solver_ptr = new Block( linearSolver ); // 矩阵块求解器
// 梯度下降方法,从GN, LM, DogLeg 中选
g2o::OptimizationAlgorithmLevenberg* solver = new g2o::OptimizationAlgorithmLevenberg( solver_ptr );//6.2.3 Levenberg-Marquadt
// g2o::OptimizationAlgorithmGaussNewton* solver = new g2o::OptimizationAlgorithmGaussNewton( solver_ptr );
// g2o::OptimizationAlgorithmDogleg* solver = new g2o::OptimizationAlgorithmDogleg( solver_ptr );
g2o::SparseOptimizer optimizer; // 图模型
optimizer.setAlgorithm( solver ); // 设置求解器
optimizer.setVerbose( true ); // 打开调试输出
// 往图中增加顶点
CurveFittingVertex* v = new CurveFittingVertex();
v->setEstimate( Eigen::Vector3d(0,0,0) );
v->setId(0);
optimizer.addVertex( v );
// 往图中增加边
for ( int i=0; i<N; i++ )
{
CurveFittingEdge* edge = new CurveFittingEdge( x_data[i] );
edge->setId(i);
edge->setVertex( 0, v ); // 设置连接的顶点,0是 v->setId(0) ??
edge->setMeasurement( y_data[i] ); // 观测数值
edge->setInformation( Eigen::Matrix<double,1,1>::Identity()*1/(w_sigma*w_sigma) ); // 信息矩阵:协方差矩阵之逆 ???
optimizer.addEdge( edge );
}
// 执行优化
cout<<"start optimization"<<endl;
chrono::steady_clock::time_point t1 = chrono::steady_clock::now();
optimizer.initializeOptimization();
optimizer.optimize(100);
chrono::steady_clock::time_point t2 = chrono::steady_clock::now();
chrono::duration<double> time_used = chrono::duration_cast<chrono::duration<double>>( t2-t1 );
cout<<"solve time cost = "<<time_used.count()<<" seconds. "<<endl;
// 输出优化值
Eigen::Vector3d abc_estimate = v->estimate();
cout<<"estimated model: "<<abc_estimate.transpose()<<endl;
return 0;
}
// 梯度下降方法,从GN, LM, DogLeg 中选
g2o::OptimizationAlgorithmLevenberg* solver = new g2o::OptimizationAlgorithmLevenberg( solver_ptr );//6.2.3 Levenberg-Marquadt
// g2o::OptimizationAlgorithmGaussNewton* solver = new g2o::OptimizationAlgorithmGaussNewton( solver_ptr );
// g2o::OptimizationAlgorithmDogleg* solver = new g2o::OptimizationAlgorithmDogleg( solver_ptr );
solve time cost = 0.000989289 seconds.
estimated model: 0.890912 2.1719 0.943629
solve time cost = 0.00265157 seconds.
estimated model: 394.439 -169.761 3.84695
solve time cost = 0.00105172 seconds.
estimated model: 0.890911 2.1719 0.943629
BA优化 PnP 的结果示例2
我们来尝试RANSAC PnP 加上迭代优化的方式估计相机位姿,看看是否对前一节的效果有所改进。非线性优化问题的求解。由于本节的目标是估计位姿而非结构,我们以相机位姿 ξ 为优化变量,通过最小化重投影误差,来构建优化问题。与之前一样,我们自定义一个 g2o 中的优化边。它只优化一个位姿,因此是一个一元边。
由之前的PnP,可以求出一个R,t,K。而且空间点的世界坐标知道,第二个相机位姿的像素坐标也是知道的。就可以利用它们进行优化。
首先确定变量为const vector<Point3f> points_3d,const vector<Point2f>,const Mat& K,Mat& R,Mat& t.
因为之后放进去的pts_3d,pts_2d是我们自己计算出来的,所以不用用&。
开始函数:
1.初始化g2o.这个可以是固定的方式。基本上都是相似的。
定义矩阵块的类型为Block.这样方便之后的写入。它有两个参数,第一个参数为优化变量的维度,这里为6,第二个参数为误差值的维度,这里为3。
步骤就是,先把复杂的矩阵块类型在这里定义为Block.然后定义线性方程求解器linearsolver,再定义矩阵块求解器solver_ptr和linearsolver有关,再定义梯度下降方法solver和solver_ptr有关。
// using bundle adjustment to optimize the pose
//步骤就是,先把复杂的矩阵块类型在这里定义为Block.然后定义线性方程求解器linearsolver,再定义矩阵块求解器solver_ptr和linearsolver有关,再定义梯度下降方法solver和solver_ptr有关。
//定义矩阵块的类型为Block.这样方便之后的写入。它有两个参数,第一个参数为优化变量的维度,这里为6,第二个参数为误差值的维度,这里为2。
typedef g2o::BlockSolver<g2o::BlockSolverTraits<6,2>> Block;
//定义线性求解器的类型linearsolver要带指针,有稠密Dense和CSparse两种类型。前面要加new.参数为矩阵块里的位姿矩阵类型
Block::LinearSolverType* linearSolver = new g2o::LinearSolverDense<Block::PoseMatrixType>();
//Block::LinearSolverType* linearSolver = new g2o::LinearSolverCSparse<Block::PoseMatrixType>();//XXX
//定义矩阵块的求解器solver_ptr。要带指针。也要带new.g2o赋初值的都要带new.
Block* solver_ptr = new Block( linearSolver );
//定义梯度下降方法为levenberg方法solver.
g2o::OptimizationAlgorithmLevenberg* solver = new g2o::OptimizationAlgorithmLevenberg ( solver_ptr );
//定义图模型optimizer.类型为g2o::SparseOptimizer.
g2o::SparseOptimizer optimizer;
//然后设置图模型的求解器,也就是之前定义的梯度下降方法。
optimizer.setAlgorithm ( solver );
// // 设置要不要打开调试输出,设置为true.这个选项有的时候可以不设
// optimizer.setVerbose(true);
2.设置顶点有关的东西,这里有两个顶点,一个是李代数位姿pose,一个是空间点位置p。
//首先确定变量为const vector<Point3f> points_3d,const vector<Point2f>,const Mat& K,Mat& R,Mat& t.
//因为之后放进去的pts_3d,pts_2d是我们自己计算出来的,所以不用 用&
void VisualOdometry::poseEstimationPnP()
{
// construct the 3d 2d observations
vector<cv::Point3f> pts3d;
vector<cv::Point2f> pts2d;
for ( cv::DMatch m:feature_matches_ )
{
pts3d.push_back( pts_3d_ref_[m.queryIdx] );
pts2d.push_back( keypoints_curr_[m.trainIdx].pt );
}
Mat K = ( cv::Mat_<double>(3,3)<<
ref_->camera_->fx_, 0, ref_->camera_->cx_,
0, ref_->camera_->fy_, ref_->camera_->cy_,
0,0,1
);
Mat rvec, tvec, inliers;
cv::solvePnPRansac( pts3d, pts2d, K, Mat(), rvec, tvec, false, 100, 4.0, 0.99, inliers );
num_inliers_ = inliers.rows;
cout<<"pnp inliers: "<<num_inliers_<<endl;
//优化 PnP 的结果 将优pnpRansac的初值给T_c_r_estimated
//一个是Estimate,这里是由R,t组成的SE3Quat形式。只要把R,t放进去就可以了。
//但是R必须是矩阵形式。由R.at<double>(0,0)等转化。t必须是向量,由t.at<double>(0,0)等转化。
T_c_r_estimated_ = SE3(
SO3(rvec.at<double>(0,0), rvec.at<double>(1,0), rvec.at<double>(2,0)),
Vector3d( tvec.at<double>(0,0), tvec.at<double>(1,0), tvec.at<double>(2,0))
);
// using bundle adjustment to optimize the pose
/*初始化g2o.这个可以是固定的方式*/
//步骤就是,先把复杂的矩阵块类型在这里定义为Block.然后定义线性方程求解器linearsolver,再定义矩阵块求解器solver_ptr和linearsolver有关,再定义梯度下降方法solver和solver_ptr有关。
//定义矩阵块的类型为Block.这样方便之后的写入。它有两个参数,第一个参数为优化变量的维度,这里为6,第二个参数为误差值的维度,这里为2。
typedef g2o::BlockSolver<g2o::BlockSolverTraits<6,2>> Block;
//定义线性求解器的类型linearsolver要带指针,有稠密Dense和CSparse两种类型。前面要加new.参数为矩阵块里的位姿矩阵类型
Block::LinearSolverType* linearSolver = new g2o::LinearSolverDense<Block::PoseMatrixType>();
//Block::LinearSolverType* linearSolver = new g2o::LinearSolverCSparse<Block::PoseMatrixType>();//XXX
//定义矩阵块的求解器solver_ptr。要带指针。也要带new.g2o赋初值的都要带new.
Block* solver_ptr = new Block( linearSolver );
//定义梯度下降方法为levenberg方法solver.
g2o::OptimizationAlgorithmLevenberg* solver = new g2o::OptimizationAlgorithmLevenberg ( solver_ptr );
//定义图模型optimizer.类型为g2o::SparseOptimizer.
g2o::SparseOptimizer optimizer;
//然后设置图模型的求解器,也就是之前定义的梯度下降方法。
optimizer.setAlgorithm ( solver );
// // 设置要不要打开调试输出,设置为true.这个选项有的时候可以不设
// optimizer.setVerbose(true);
/*设置顶点有关的东西,这里有两个顶点,一个是李代数位姿pose,一个是空间点位置p。*/
//李代数位姿顶点类型为g2o::VertexSE3Expmap*,初值为new g2o::VertexSE3Expmap();
g2o::VertexSE3Expmap* pose = new g2o::VertexSE3Expmap();
//顶点要设置两个东西,一个是Id,一般设为0,
pose->setId ( 0 );
//一个是Estimate,这里是由R,t组成的SE3Quat形式。只要把R,t放进去就可以了。
//但是R必须是矩阵形式。由之前的R.at<double>(0,0)等转化。t必须是向量,由t.at<double>(0,0)等转化。
pose->setEstimate ( g2o::SE3Quat (
T_c_r_estimated_.rotation_matrix(),
T_c_r_estimated_.translation()
) );
optimizer.addVertex ( pose );
// edges
//边肯定在一个for循环里,边的类型为g2o::EdgeProjectXYZ2UV,初值为new g2o::EdgeProjectXYZ2UV().
for ( int i=0; i<inliers.rows; i++ )
{
int index = inliers.at<int>(i,0);
// 3D -> 2D projection
EdgeProjectXYZ2UVPoseOnly* edge = new EdgeProjectXYZ2UVPoseOnly();
//设置id为index,
edge->setId(i);
//设置顶点0为位姿。
edge->setVertex(0, pose);
//相机参数camera.类型为g2o::CameraParmetersI,值为K.at<double>(0,0),和cx,cy组成的2维向量,
edge->camera_ = curr_->camera_.get();
edge->point_ = Vector3d( pts3d[index].x, pts3d[index].y, pts3d[index].z );
edge->setMeasurement( Vector2d(pts2d[index].x, pts2d[index].y) );
edge->setInformation( Eigen::Matrix2d::Identity() );
optimizer.addEdge( edge );
}
//求解
//求解就特别简单,先把图模型给初始化一下。
optimizer.initializeOptimization();
//直接用优化函数optimize()进行优化
optimizer.optimize(10);
//由优化可以得到位姿的估计值
//pose->estimate.要想把它转成常见形式,可以用欧式变换矩阵Eigen::Isometry3d,是4*4矩阵,T=Eigen::Isometry3d(pose->estimate()).matrix
T_c_r_estimated_ = SE3 (
pose->estimate().rotation(),
pose->estimate().translation()
);
}
#ifndef MYSLAM_G2O_TYPES_H
#define MYSLAM_G2O_TYPES_H
#include "myslam/common_include.h"
#include "camera.h"
#include <g2o/core/base_vertex.h>
#include <g2o/core/base_unary_edge.h>
#include <g2o/core/block_solver.h>
#include <g2o/core/optimization_algorithm_levenberg.h>
#include <g2o/types/sba/types_six_dof_expmap.h>
#include <g2o/solvers/dense/linear_solver_dense.h>
#include <g2o/core/robust_kernel.h>
#include <g2o/core/robust_kernel_impl.h>
namespace myslam
{
class EdgeProjectXYZRGBD : public g2o::BaseBinaryEdge<3, Eigen::Vector3d, g2o::VertexSBAPointXYZ, g2o::VertexSE3Expmap>
{
public:
EIGEN_MAKE_ALIGNED_OPERATOR_NEW;
virtual void computeError();
virtual void linearizeOplus();
virtual bool read( std::istream& in ){}
virtual bool write( std::ostream& out) const {}
};
//李代数位姿顶点类型为g2o::VertexSE3Expmap*,初值为new g2o::VertexSE3Expmap();
// only to optimize the pose, no point
class EdgeProjectXYZRGBDPoseOnly: public g2o::BaseUnaryEdge<3, Eigen::Vector3d, g2o::VertexSE3Expmap >
{
public:
EIGEN_MAKE_ALIGNED_OPERATOR_NEW
// Error: measure = R*point+t
virtual void computeError();
virtual void linearizeOplus();
virtual bool read( std::istream& in ){}
virtual bool write( std::ostream& out) const {}
Vector3d point_;
};
//李代数位姿顶点类型为g2o::VertexSE3Expmap
class EdgeProjectXYZ2UVPoseOnly: public g2o::BaseUnaryEdge<2, Eigen::Vector2d, g2o::VertexSE3Expmap >
{
public:
EIGEN_MAKE_ALIGNED_OPERATOR_NEW
virtual void computeError();
virtual void linearizeOplus();
virtual bool read( std::istream& in ){}
virtual bool write(std::ostream& os) const {};
Vector3d point_;
Camera* camera_;
};
}
#endif // MYSLAM_G2O_TYPES_H
#include "myslam/g2o_types.h"
//以相机位姿 ξ 为优化变量,通过最小化重投影误差,来构建优化问题
namespace myslam
{
void EdgeProjectXYZRGBD::computeError()
{
const g2o::VertexSBAPointXYZ* point = static_cast<const g2o::VertexSBAPointXYZ*> ( _vertices[0] );
const g2o::VertexSE3Expmap* pose = static_cast<const g2o::VertexSE3Expmap*> ( _vertices[1] );
_error = _measurement - pose->estimate().map ( point->estimate() );
}
void EdgeProjectXYZRGBD::linearizeOplus()
{
g2o::VertexSE3Expmap* pose = static_cast<g2o::VertexSE3Expmap *> ( _vertices[1] );
g2o::SE3Quat T ( pose->estimate() );
g2o::VertexSBAPointXYZ* point = static_cast<g2o::VertexSBAPointXYZ*> ( _vertices[0] );
Eigen::Vector3d xyz = point->estimate();
Eigen::Vector3d xyz_trans = T.map ( xyz );
double x = xyz_trans[0];
double y = xyz_trans[1];
double z = xyz_trans[2];
_jacobianOplusXi = - T.rotation().toRotationMatrix();
_jacobianOplusXj ( 0,0 ) = 0;
_jacobianOplusXj ( 0,1 ) = -z;
_jacobianOplusXj ( 0,2 ) = y;
_jacobianOplusXj ( 0,3 ) = -1;
_jacobianOplusXj ( 0,4 ) = 0;
_jacobianOplusXj ( 0,5 ) = 0;
_jacobianOplusXj ( 1,0 ) = z;
_jacobianOplusXj ( 1,1 ) = 0;
_jacobianOplusXj ( 1,2 ) = -x;
_jacobianOplusXj ( 1,3 ) = 0;
_jacobianOplusXj ( 1,4 ) = -1;
_jacobianOplusXj ( 1,5 ) = 0;
_jacobianOplusXj ( 2,0 ) = -y;
_jacobianOplusXj ( 2,1 ) = x;
_jacobianOplusXj ( 2,2 ) = 0;
_jacobianOplusXj ( 2,3 ) = 0;
_jacobianOplusXj ( 2,4 ) = 0;
_jacobianOplusXj ( 2,5 ) = -1;
}
void EdgeProjectXYZRGBDPoseOnly::computeError()
{
const g2o::VertexSE3Expmap* pose = static_cast<const g2o::VertexSE3Expmap*> ( _vertices[0] );
_error = _measurement - pose->estimate().map ( point_ );
}
void EdgeProjectXYZRGBDPoseOnly::linearizeOplus()
{
g2o::VertexSE3Expmap* pose = static_cast<g2o::VertexSE3Expmap*> ( _vertices[0] );
g2o::SE3Quat T ( pose->estimate() );
Vector3d xyz_trans = T.map ( point_ );
double x = xyz_trans[0];
double y = xyz_trans[1];
double z = xyz_trans[2];
_jacobianOplusXi ( 0,0 ) = 0;
_jacobianOplusXi ( 0,1 ) = -z;
_jacobianOplusXi ( 0,2 ) = y;
_jacobianOplusXi ( 0,3 ) = -1;
_jacobianOplusXi ( 0,4 ) = 0;
_jacobianOplusXi ( 0,5 ) = 0;
_jacobianOplusXi ( 1,0 ) = z;
_jacobianOplusXi ( 1,1 ) = 0;
_jacobianOplusXi ( 1,2 ) = -x;
_jacobianOplusXi ( 1,3 ) = 0;
_jacobianOplusXi ( 1,4 ) = -1;
_jacobianOplusXi ( 1,5 ) = 0;
_jacobianOplusXi ( 2,0 ) = -y;
_jacobianOplusXi ( 2,1 ) = x;
_jacobianOplusXi ( 2,2 ) = 0;
_jacobianOplusXi ( 2,3 ) = 0;
_jacobianOplusXi ( 2,4 ) = 0;
_jacobianOplusXi ( 2,5 ) = -1;
}
void EdgeProjectXYZ2UVPoseOnly::computeError()
{
const g2o::VertexSE3Expmap* pose = static_cast<const g2o::VertexSE3Expmap*> ( _vertices[0] );
_error = _measurement - camera_->camera2pixel (
pose->estimate().map(point_) );
}
void EdgeProjectXYZ2UVPoseOnly::linearizeOplus()
{
g2o::VertexSE3Expmap* pose = static_cast<g2o::VertexSE3Expmap*> ( _vertices[0] );
g2o::SE3Quat T ( pose->estimate() );
Vector3d xyz_trans = T.map ( point_ );
double x = xyz_trans[0];
double y = xyz_trans[1];
double z = xyz_trans[2];
double z_2 = z*z;
_jacobianOplusXi ( 0,0 ) = x*y/z_2 *camera_->fx_;
_jacobianOplusXi ( 0,1 ) = - ( 1+ ( x*x/z_2 ) ) *camera_->fx_;
_jacobianOplusXi ( 0,2 ) = y/z * camera_->fx_;
_jacobianOplusXi ( 0,3 ) = -1./z * camera_->fx_;
_jacobianOplusXi ( 0,4 ) = 0;
_jacobianOplusXi ( 0,5 ) = x/z_2 * camera_->fx_;
_jacobianOplusXi ( 1,0 ) = ( 1+y*y/z_2 ) *camera_->fy_;
_jacobianOplusXi ( 1,1 ) = -x*y/z_2 *camera_->fy_;
_jacobianOplusXi ( 1,2 ) = -x/z *camera_->fy_;
_jacobianOplusXi ( 1,3 ) = 0;
_jacobianOplusXi ( 1,4 ) = -1./z *camera_->fy_;
_jacobianOplusXi ( 1,5 ) = y/z_2 *camera_->fy_;
}
}
李代数位姿顶点类型为g2o::VertexSE3Expmap*,初值为new g2o::VertexSE3Expmap();
顶点要设置两个东西,一个是Id,一般设为0,一个是Estimate,这里是由R,t组成的SE3Quat形式。只要把R,t放进去就可以了。但是R必须是矩阵形式。由之前的R.at<double>(0,0)等转化。t必须是向量,由t.at<double>(0,0)等转化。
至于空间点式的顶点,就需要一个for循环了,因为空间点有很多啊。
for(const Point3f p:points_3d)
顶点设为point,类型为g2o::VertexSBAPointXYZ,初值为new g2o::VertexSBAPointXYZ();
Id设为index++,估计值设为p.x,p.y,p.z的组成的向量形式。还设置了一个setMarginalized(true).
3相机参数
里面还用到了相机参数camera.类型为g2o::CameraParmetersI,值为K.at<double>(0,0),和cx,cy组成的2维向量,0组成的。
Id设置为0,
4.边。
重点是边了。边肯定在一个for循环里,这里是for(const Point2f p:points_2d)
边的类型为g2o::EdgeProjectXYZ2UV,初值为new g2o::EdgeProjectXYZ2UV().
设置id为index,
设置顶点0为空间点位置,而且里面用了一个dyanmic_cast把图模型的顶点坐标设置为空间点顶点类型。
edge->setVertex(0,dynamic_cat<g2o::VertexSBAPointXYZ*>(optimizer.vertex(index)));
设置顶点1为位姿。
edge->setVertex(1,pose);
设置测量值为p.x,p.y组成的2维向量模式。
设置参数ID为(0,0).
edge->setMeasurement(Eigen::Vector2d(p.x,p.y));
edge->setParameterId(0,0);
设置信息矩阵为2维的单位矩阵。信息矩阵的维度是根据误差值的维度定的。
5.求解
求解就特别简单,先把图模型给初始化一下。
optimizer.IntializeOptimization();
直接用优化函数optimize()进行优化。
optimizer.optimize(100)
由优化可以得到位姿的估计值。pose->estimate.要想把它转成常见形式,可以用欧式变换矩阵Eigen::Isometry3d,是4*4矩阵,
T=Eigen::Isometry3d(pose->estimate()).matrix
初始化g2o.这个可以是固定的方式。基本上都是相似的。
//顶点
// 往图中增加边
说明:
//定义矩阵块的类型为Block.这样方便之后的写入。它有两个参数,第一个参数为优化变量的维度,这里为6(R,T),第二个参数为误差值的维度,这里为2(重投影x,y 没以和表示吗)。//3
typedef g2o::BlockSolver<g2o::BlockSolverTraits<6,1>> Block;//6,1 6,2 6,3
其中BlockSolverTraits<6,1>> Block 可以为6,1; 6,2;6,3结果相同. 如果为5 报错,如果为7可以,但结果稍微不同
稀疏直接法
计算光度误差的边,它的雅可比矩阵
g20
bool poseEstimationDirect ( const vector< Measurement >& measurements, cv::Mat* gray, Eigen::Matrix3f& K, Eigen::Isometry3d& Tcw )
{
// 初始化g2o
typedef g2o::BlockSolver<g2o::BlockSolverTraits<6,1>> DirectBlock; // 求解的向量是6*1的
DirectBlock::LinearSolverType* linearSolver = new g2o::LinearSolverDense< DirectBlock::PoseMatrixType > ();
DirectBlock* solver_ptr = new DirectBlock ( linearSolver );
// g2o::OptimizationAlgorithmGaussNewton* solver = new g2o::OptimizationAlgorithmGaussNewton( solver_ptr ); // G-N
g2o::OptimizationAlgorithmLevenberg* solver = new g2o::OptimizationAlgorithmLevenberg ( solver_ptr ); // L-M
g2o::SparseOptimizer optimizer;
optimizer.setAlgorithm ( solver );
optimizer.setVerbose( true );
g2o::VertexSE3Expmap* pose = new g2o::VertexSE3Expmap();
pose->setEstimate ( g2o::SE3Quat ( Tcw.rotation(), Tcw.translation() ) );
pose->setId ( 0 );
optimizer.addVertex ( pose );
// 添加边
int id=1;
for ( Measurement m: measurements )
{
EdgeSE3ProjectDirect* edge = new EdgeSE3ProjectDirect (
m.pos_world,
K ( 0,0 ), K ( 1,1 ), K ( 0,2 ), K ( 1,2 ), gray
);
edge->setVertex ( 0, pose );
edge->setMeasurement ( m.grayscale );
edge->setInformation ( Eigen::Matrix<double,1,1>::Identity() );
edge->setId ( id++ );
optimizer.addEdge ( edge );
}
cout<<"edges in graph: "<<optimizer.edges().size() <<endl;
optimizer.initializeOptimization();
optimizer.optimize ( 30 );
Tcw = pose->estimate();
}
雅可比矩阵
// project a 3d point into an image plane, the error is photometric error
// an unary edge with one vertex SE3Expmap (the pose of camera)
class EdgeSE3ProjectDirect: public BaseUnaryEdge< 1, double, VertexSE3Expmap>
{
public:
EIGEN_MAKE_ALIGNED_OPERATOR_NEW
EdgeSE3ProjectDirect() {}
EdgeSE3ProjectDirect ( Eigen::Vector3d point, float fx, float fy, float cx, float cy, cv::Mat* image )
: x_world_ ( point ), fx_ ( fx ), fy_ ( fy ), cx_ ( cx ), cy_ ( cy ), image_ ( image )
{}
virtual void computeError()
{
const VertexSE3Expmap* v =static_cast<const VertexSE3Expmap*> ( _vertices[0] );
Eigen::Vector3d x_local = v->estimate().map ( x_world_ );
float x = x_local[0]*fx_/x_local[2] + cx_;
float y = x_local[1]*fy_/x_local[2] + cy_;
// check x,y is in the image
if ( x-4<0 || ( x+4 ) >image_->cols || ( y-4 ) <0 || ( y+4 ) >image_->rows )
{
_error ( 0,0 ) = 0.0;
this->setLevel ( 1 );
}
else
{
_error ( 0,0 ) = getPixelValue ( x,y ) - _measurement;
}
}
// plus in manifold
virtual void linearizeOplus( )
{
if ( level() == 1 )
{
_jacobianOplusXi = Eigen::Matrix<double, 1, 6>::Zero();
return;
}
VertexSE3Expmap* vtx = static_cast<VertexSE3Expmap*> ( _vertices[0] );
Eigen::Vector3d xyz_trans = vtx->estimate().map ( x_world_ ); // q in book _r*xyz + _t;
double x = xyz_trans[0];
double y = xyz_trans[1];
double invz = 1.0/xyz_trans[2];
double invz_2 = invz*invz;
float u = x*fx_*invz + cx_;
float v = y*fy_*invz + cy_;
// jacobian from se3 to u,v
// NOTE that in g2o the Lie algebra is (\omega, \epsilon), where \omega is so(3) and \epsilon the translation
Eigen::Matrix<double, 2, 6> jacobian_uv_ksai;
jacobian_uv_ksai ( 0,0 ) = - x*y*invz_2 *fx_;
jacobian_uv_ksai ( 0,1 ) = ( 1+ ( x*x*invz_2 ) ) *fx_;
jacobian_uv_ksai ( 0,2 ) = - y*invz *fx_;
jacobian_uv_ksai ( 0,3 ) = invz *fx_;
jacobian_uv_ksai ( 0,4 ) = 0;
jacobian_uv_ksai ( 0,5 ) = -x*invz_2 *fx_;
jacobian_uv_ksai ( 1,0 ) = - ( 1+y*y*invz_2 ) *fy_;
jacobian_uv_ksai ( 1,1 ) = x*y*invz_2 *fy_;
jacobian_uv_ksai ( 1,2 ) = x*invz *fy_;
jacobian_uv_ksai ( 1,3 ) = 0;
jacobian_uv_ksai ( 1,4 ) = invz *fy_;
jacobian_uv_ksai ( 1,5 ) = -y*invz_2 *fy_;
Eigen::Matrix<double, 1, 2> jacobian_pixel_uv;
jacobian_pixel_uv ( 0,0 ) = ( getPixelValue ( u+1,v )-getPixelValue ( u-1,v ) ) /2;
jacobian_pixel_uv ( 0,1 ) = ( getPixelValue ( u,v+1 )-getPixelValue ( u,v-1 ) ) /2;
_jacobianOplusXi = jacobian_pixel_uv*jacobian_uv_ksai;
}
// dummy read and write functions because we don't care...
virtual bool read ( std::istream& in ) {}
virtual bool write ( std::ostream& out ) const {}
protected:
// get a gray scale value from reference image (bilinear interpolated)
inline float getPixelValue ( float x, float y )
{
uchar* data = & image_->data[ int ( y ) * image_->step + int ( x ) ];
float xx = x - floor ( x );
float yy = y - floor ( y );
return float (
( 1-xx ) * ( 1-yy ) * data[0] +
xx* ( 1-yy ) * data[1] +
( 1-xx ) *yy*data[ image_->step ] +
xx*yy*data[image_->step+1]
);
}
public:
Eigen::Vector3d x_world_; // 3D point in world frame
float cx_=0, cy_=0, fx_=0, fy_=0; // Camera intrinsics
cv::Mat* image_=nullptr; // reference image
};