gtsam  4.1.0
gtsam
Geometry

The most common 5DOF 3D->2D calibration. More...

The most common 5DOF 3D->2D calibration.

A 2D stereo point, v will be same for rectified images.

A stereo camera class, parameterize by left camera pose and stereo calibration.

A 3D rotation represented as a rotation matrix if the preprocessor symbol GTSAM_USE_QUATERNIONS is not defined, or as a quaternion if it is defined.

Rotation matrix NOTE: the angle theta is in radians unless explicitly stated.

A 3D pose (R,t) : (Rot3,Point3)

A 2D pose (Point2,Rot2)

A pinhole camera class that has a Pose3 and a fixed Calibration.

A pinhole camera class that has a Pose3 and a Calibration.

A 3D line (R,a,b) : (Rot3,Scalar,Scalar)

A Calibrated camera class [R|-R't], calibration K=I.

A pinhole camera class that has a Pose3, functions as base class for all pinhole cameras.

Calibration of a omni-directional camera with mirror + lens radial distortion.

Calibration of a fisheye camera.

Calibration of a camera with radial distortion.

Calibration of a camera with radial distortion that also supports Lie-group behaviors for optimization.

Calibration used by Bundler.

The most common 5DOF 3D->2D calibration, stereo version.

See also
Cal3DS2_Base

Uses same distortionmodel as OpenCV, with http://docs.opencv.org/modules/calib3d/doc/camera_calibration_and_3d_reconstruction.html but using only k1,k2,p1, and p2 coefficients. K = [ fx s u0 ; 0 fy v0 ; 0 0 1 ] rr = Pn.x^2 + Pn.y^2 \hat{Pn} = (1 + k1*rr + k2*rr^2 ) Pn + [ 2*p1 Pn.x Pn.y + p2 (rr + 2 Pn.x^2) ; p1 (rr + 2 Pn.y^2) + 2*p2 Pn.x Pn.y ] pi = K*Pn

Uses same distortionmodel as OpenCV, with https://docs.opencv.org/master/db/d58/group__calib3d__fisheye.html 3D point in camera frame p = (x, y, z) Intrinsic coordinates: [x_i;y_i] = [x/z; y/z] Distorted coordinates: r^2 = (x_i)^2 + (y_i)^2 th = atan(r) th_d = th(1 + k1*th^2 + k2*th^4 + k3*th^6 + k4*th^8) [x_d; y_d] = (th_d / r)*[x_i; y_i] Pixel coordinates: K = [fx s u0; 0 fy v0 ;0 0 1] [u; v; 1] = K*[x_d; y_d; 1]

Similar to Cal3DS2, does distortion but has additional mirror parameter xi K = [ fx s u0 ; 0 fy v0 ; 0 0 1 ] Pn = [ P.x / (1 + xi * \sqrt{P.x^2 + P.y^2 + 1}), P.y / (1 + xi * \sqrt{P.x^2 + P.y^2 + 1})] rr = Pn.x^2 + Pn.y^2 \hat{pn} = (1 + k1*rr + k2*rr^2 ) pn + [ 2*k3 pn.x pn.y + k4 (rr + 2 Pn.x^2) ; k3 (rr + 2 Pn.y^2) + 2*k4 pn.x pn.y ] pi = K*pn

If calibration is known, it is more computationally efficient to calibrate the measurements rather than try to predict in pixels.

Use PinholePose if you will not be optimizing for Calibration

Instead of using this class, one might consider calibrating the measurements and using CalibratedCamera, which would then be faster.