DLR CalDe and CalLab -
The DLR Camera Calibration Toolbox

DLR CalDe and DLR CalLab emerged late in the year 2005 at the Institute of Robotics and Mechatronics. They followed from the strategic purpose of both upgrading the former CalLab package and at the same time developing a platform independent application.

Having platform independency in mind, it was chosen to develop in the IDL language. In addition to that, this choice yielded reduced development time and boosted performance. Thanks to the IDL Virtual Machine concept, the toolbox can be used by every interested researcher without any additional costs.

In respect of the application design it was decided to detach the features detection procedures from the application for parameters estimation. The former task is now performed by the program DLR CalDe, which is completely independent of DLR CalLab, whereas the latter task is exclusively performed by DLR CalLab.

The following images give a first impression of the look-and-feel of the calibration toolbox.

DLR CalDe (DLR Calibration Detection Toolbox)

The detection toolbox DLR CalDe serves the need for localizing landmarks/corners on a chessboard-like 2D calibration panel with sub-pixel accuracy.

In contrast to the vast majority of similar free available toolboxes, here the operation is fully automatic. In addition to that, the calibration pattern no longer has to be fully visible within the images. Implications of this fact are twofold: First, makes it possible to successfully estimate the lens distortion parameters also in the peripheral regions of the image. Second, facilitates the calibration of stereo-cameras and eye-in-hand or eye-to-hand systems, since partially visible patterns suffice for calibration.

However, the application also preserves the possibility of manual interaction and adjustment of the selected landmarks.

Finally, the files containing both the coordinates of the landmarks in the calibration object and their corresponding (stereo) image coordinates can be generated. These are the starting point for the camera calibration tool DLR CalLab.

DLR CalLab (DLR Calibration Laboratory)

DLR CalLab estimates both the intrinsic and the extrinsic parameters of either a single camera or a stereo-camera (i.e. of a constellation of two or more cameras rigidly attached between themselves), on the basis of the previously detected image features (e.g. from DLR CalDe).

The intrinsic parameters parameterize the camera models (pin-hole camera model and lens distortion model), thus describing the perspective projection, the optical distortions, as well as the discretization process. These define therefore the nonlinear transformation between the projection rays in camera frame and their corresponding projections in the image memory frame. In the case of a stereo-camera calibration, the rigid-body transformation(s) between cameras might also be considered as an intrinsic parameter of the stereo-camera system.

The extrinsic parameters parameterize the rigid-body transformations between the (main) camera frame and either the world frame or the Tool Centre Point (TCP) frame of a robotic manipulator. The former transformation changes at different instants (camera stations), whereas the latter remains constant as long as the camera(s) stay rigidly attached to the TCP. For the camera to TCP calibration (generally known as hand-eye calibration) several methods were implemented - refer to the work of Strobl and Hirzinger.

Furthermore, the program offers extensive interaction possibilities:

• Choice and parameterization of different numerical optimization algorithms

• Hands-on histograms and images for the selection of features to remove. A roughly accurate parameterization of a camera model can be readily obtained even out from erroneous image-to-object source data. It turns out that the predicted corners projections resulting from this estimation prove to be of good use to promptly verify the source data (by e.g. DLR CalDe).

• A variety of estimation methods both for intrinsic and for extrinsic calibration.

• Possibility to fix several intrinsic parameters (and even all of them), and to accurately determine the camera-to-object transformations corresponding to the images used.

• Flexibility in the selection of the lens distortion model - if any - up to radial distortion in 3rd, 5th, and 7th orders, decentering distortion in 2nd and 4th orders, and thin prism distortion in 2nd and 4th orders.

Nevertheless, it is also possible to perform automatically the whole calibration in one-button-mode.

Corresponding authors: Klaus Strobl, Wolfgang Sepp, Stefan Fuchs, and Klaus Arbter; DLR RM
www.robotic.dlr.de/callab/