mpkuse/camera_model

part of camodocal. Originally developed by Lionel Heng.

Required

How to compile (Standalone)

mkdir build
cmake ..
make

It is very easy to adapt this into your own projects. You will need to copy the include and src directories to your project and edit your cmake files accordingly. Look at my cmake file to know how to do it.

Calibration (Mono):

Use intrinsic_calib.cc to calibrate your camera. You may use github-krosutils to get images out of a bag recording.

Go here to get printable checkerboards of various dimensions.

$ ./Calibration -w 9 -h 6 -s 35 -i ~/try/left_images -p image -e .png --camera-name leftcamera --view-results

$ ./Calibration --help
Allowed options:
  --help                                produce help message
  -w [ --width ] arg (=8)               Number of inner corners on the
                                        chessboard pattern in x direction
  -h [ --height ] arg (=12)             Number of inner corners on the
                                        chessboard pattern in y direction
  -s [ --size ] arg (=7)                Size of one square in mm
  -i [ --input ] arg (=calibrationdata) Input directory containing chessboard
                                        images
  -p [ --prefix ] arg (=left-)          Prefix of images
  -e [ --file-extension ] arg (=.png)   File extension of images
  --camera-model arg (=mei)             Camera model: kannala-brandt | mei |
                                        pinhole
  --camera-name arg (=camera)           Name of camera
  --opencv                              Use OpenCV to detect corners
  --view-results                        View results
  -v [ --verbose ]                      Verbose output

Calibration (Stereo):

$ ./Calibration_Stereo -w 9 -h 6 -s 37.4 -i ~/try/sync_stereo_pairs/ --prefix-l leftimage- --prefix-r rightimage- -e .png --view-result

$ ./Calibration_Stereo --help
Allowed options:
  --help                              produce help message
  -w [ --width ] arg (=9)             Number of inner corners on the chessboard
                                      pattern in x direction
  -h [ --height ] arg (=6)            Number of inner corners on the chessboard
                                      pattern in y direction
  -s [ --size ] arg (=120)            Size of one square in mm
  -i [ --input ] arg (=images)        Input directory containing chessboard
                                      images
  -o [ --output ] arg (=.)            Output directory containing calibration
                                      data
  --prefix-l arg (=left)              Prefix of images from left camera
  --prefix-r arg (=right)             Prefix of images from right camera
  -e [ --file-extension ] arg (=.bmp) File extension of images
  --camera-model arg (=mei)           Camera model: kannala-brandt | mei |
                                      pinhole
  --camera-name-l arg (=camera_left)  Name of left camera
  --camera-name-r arg (=camera_right) Name of right camera
  --opencv                            Use OpenCV to detect corners
  --view-results                      View results
  -v [ --verbose ]                    Verbose output

Undistortion / Geometry:

See Camera.h for general interface:

USE:

Two main files for you to use camera model: Camera.h and CameraFactory.h.

1.load in the camera model calibration file

Use function in CameraFactory.h to load in the camra calibration file:

#include <camera_model/camera_models/CameraFactory.h>

camera_model::CameraPtr cam;

void loadCameraFile(std::string camera_model_file)
{
    cam = camera_model::CameraFactory::instance()->generateCameraFromYamlFile(camera_model_file);
}

void makeFromScratch( )
{
    cam = camodocal::CameraFactory::instance()->generateCamera( camodocal::Camera::PINHOLE, "my_camera", cv::Size(480,640) );
    vector<double>parameterVec;
    cam->writeParameters( parameterVec );
    parameterVec[4] = 481.20; //fx
    parameterVec[5] = 480.; //fy
    parameterVec[6] = 319.50; //cx
    parameterVec[7] = 239.50; //cy
    cam->readParameters( parameterVec );
    cout << "Cam:\n" << cam->parametersToString() << endl;

    // Similarly other camera types can also be loaded from scratch,
    // just be careful on the ordering of the parameterVec, see relavant source code for details.
}

2.projection and back-projection point

See Camera.h for general interface:

Projection (3D ---> 2D) function: spaceToPlane :Projects 3D points to the image plane (Pi function)

#include <camera_model/camera_models/CameraFactory.h>

camera_model::CameraPtr cam;

void loadCameraFile(std::string camera_model_file)
{
    cam = camera_model::CameraFactory::instance()->generateCameraFromYamlFile(camera_model_file);
}

void useProjection(Eigen::Vector3d P)
{
    Eigen::Vector2d p_dst;
    cam->spaceToPlane(P, p_dst);
}

Back Projection (2D ---> 3D) function: liftSphere: Lift points from the image plane to the projective space.

#include <camera_model/camera_models/CameraFactory.h>

camera_model::CameraPtr cam;

void loadCameraFile(std::string camera_model_file)
{
    cam = camera_model::CameraFactory::instance()->generateCameraFromYamlFile(camera_model_file);
}

void useProjection(Eigen::Vector3d P)
{
    Eigen::Vector2d p_dst;
    cam->spaceToPlane(P, p_dst);
}

void useBackProjection(Eigen::Vector2d p)
{
    Eigen::Vector3d P_dst;
    cam->liftSphere(p, P_dst);
}

liftSphere

liftProjective
normalize vector

liftProjective

K.inverse() * [ u;v;1]
apply inverse distortion

spaceToPlace (Pi function)

P := (X/Z, Y/Z)
\pho := apply distortion to P
K . \pho

lemma on spaceToPlane

=> if input to spaceToPlane is (u,v,1) instead of a real 3d point, it has the effect of applying distorion to this point, assuming the input (u,v,1) is normalized image co-ordinate

Examples

Example usage of camodocal by mpkuse.

1_readcamera_from_yaml.cpp: How to initialize CameraPtr with yaml file.
2_readcamera_from_scratch.cpp: How to initialize CameraPtr with hardcoded params

mpkuse / camera_model