yucao42/g2o_pose_optimization

Use g2o to optimize the camera pose. Initial guess is from EPnP.

Pose Optimization Using G2O Library: an Easy Example

A small sample for pose optimization using g2o to minimize reprojection errors.

Usage:

Requirement:

1. OpenCV 3.1.0 or newer version.

2. LAPACK && Suitesparse && Eigen3.

3. (optional) Viz module in opencv (needs compiling with VTK). Disable the display in the macro DISPLAY_3D.

Example:

There is an example given on optimizing a pose calculated from EPnP to give intuitions on the performance of g2o.

You may write extra code to implement other optimization based on g2o.

g2o_pose_optimization's Issues

Using a squares chessboard?

Hi, and thank you for making this code available. I realise this is an old repo, but it is exactly what I am trying to do (calculate very accurate extrinsics of two different cameras using pnp with a chessboard, and optimisation).

I use a checkerboard, however, instead of board with dots/circles.

I see in the g2o edge function:

g2o_pose_optimization/src/Optimizer.cpp

Line 184 in 59b1c34

e->Xw[0] = (float)( (i%7)* 8);//8mm each with 7 circles a row

you have:

 e->Xw[0] = (float)( (i%7)* 8);//8mm each with 7 circles a row
 e->Xw[1] = (float)(i%7?row:++row)*8;
 e->Xw[2] = (float)(0);

I assume that I need to change this. My checkerboard is:

int grid_size_x = 7;
int grid_size_y = 5;
double squareSize = 0.045; //4.5 cm

for (int y = 0; y < grid_size_y; y++)
	{
		for (int x = 0; x < grid_size_x; x++)
		{
			object_points_.push_back(cv::Point3f(x * squareSize, y * squareSize, 0));
		}
	}

How would I go about changing your code to use this board?
Aside from the findCircles to findChessboardCorners, is there anything else that I should need to adjust?

Thanks again!