The Icon for the Video Camera Calibration View

Introduction

The Video Camera Calibration plugin provides a multitude of different calibration methods.

Feature detectors: OpenCV chessboard, OpenCV assymetrical circle pattern, AprilTags, template matching of dots and rings.
Iterative refinement of control points [1].
Optimisation: 2D (reprojection) error or 3D (reconstruction) error.
Handeye calibration: Tsai 1989 [2], Shahidi 2002 [3], an adaptation of Malti 2013 [4].

Preferences

Before attempting to calibrate, you must set all the necessary preferences for your method of choice.

Figure 1: The Preference Page

previous calibration directory: You can specify a previous calibration directory and the files calib.left.intrinsic.txt, calib.right.intrinsic.txt, calib.r2l.txt, calib.left.handeye.txt, calib.right.handeye.txt and calib.model2world.txt are loaded into the plugin. This is a new and as-yet lightly tested feature, feedback is required.
rescale image: Most feature detectors run in OpenCV which ignores pixel sizes. So, if you have anisotropic pixel sizes (in millimetres), you may want to resize the image in the x and y direction.
model file: You must specifiy a model text file where each line contains the point identifier followed by x y and z coordinates all space separated.
Minimum number of views: Once you have this many views or more, calibration is recomputed at each image grab.
do 3D optimisation: If clicked, will minimise reconstruction (triangulated 3D) error, otherwise reprojection error.
do iterative calibration: See [1].
features to detect: See below.
preferred hand-eye method: This plugin will calculate 3 hand-eye methods, so you specify your "preferred" one, which means when the output is saved you get that one in calib.left.handeye.txt and when its visualised, you see your preferred one.

Feature Selection

Figure 2: Chosing the type of features to detect.

You may select different feature detectors, each of which may have different parameters to set. For example, the OpenCV chessboard requires you to set the grid size, and AprilTags requires you to set the Tag Family identifier.

Template Matching

Figure 3: Chosing the template image to match.

If doing template matching, you must provide a template of the correct scale to match.

Iterative Refinement

Figure 4: Iterative Refinement requires a canonical image, and the 2D locations of each feature. See [1].

If selected, you can do Iterative Refinement of Control Points [1], which requires a canonical (face on) image, and the exact location of each feature point in 2D.

Hand-Eye Calibration

Figure 5: Chosing the hand-eye method, see [2], [3], [4].

Note: NifTK actually uses Eye-Hand matrices, so you can just multiply a camera point by the eye-hand matrix, and then the tracker matrix to get a point in world coordinates.

Usage

Once you have set up the correct preferences, calibration is easy. Figure 6 shows the main controls.

Figure 6: The Main Controls

grab: Takes a picture, with a green tick indicating success and a red cross indicating failure.
undo: Drops the most recent picture.
clear: Removes all currently taken picture.
save: Saves all calibration information into a folder.

Once you have enough images (specified using the above preferences), calibration proceeds automatically. If the footswitch plugin is opened, you can use a USB footswitch.

Note: A known issue with the footswitch plugin is that it doesn't initialise if it is automatically open when the Workbench starts. You may need to stop and restart it.