Bachelorarbeit Ines Ben Aissa
Title:
Approaches for the geometric calibration of
thermal cameras
(Ansätze zur Kalibrierung einer
Thermographiekamera)
Description:
Cameras are calibrated to determine the geometric parameters of the interior orientation (camera constant and principal point) and the additional parameters (radial-symmetric distortion, tangential and asymmetric distortion as well as affinity and shear). The calibration of thermal cameras was mainly refrained because the users of such cameras are rather interested in the temperature resolution of the cameras, than in their geometric distortions. The importance of geometric calibration increases as the distortion towards the margin gets larger and it is expected that thermal cameras with larger sensors will be developed. The use of a reference object is problematic for the geometric calibration of a thermal camera. When a test field calibration is performed there has to be a sufficiently large thermal difference between the point markers and the test field background. Based on ideas from the literature [Luhmann et al.: 2011] a test field which matches these requirements was developed. For this a wooden board was covered with aluminum foil. Coded and un-coded points of black paper were attached (Figure 1).
The Fraunhofer Institute for Transportation and Infrastructure Systems provided three cameras with which the shots were taken. The focus used was close-up as well as an infinite focus. During the following compensations with the program Aicon 3D Studio several possible adjustments with one camera were tested to get good and representative values. In the next step, the results of the close- up range focus were compared to those of the infinite focus, with the finding that the close-up range focus provided the more accurate results. This fact can be explained through the test field which is not representative for a calibration in the infinite focus. It is too small and too close to the camera which inhibited that clear images could be taken. In a further step the individual parameters of the cameras were compared to each other. A distortion curve of the largest distortion, the radial- symmetric, has been created for all three cameras and can be seen in figure 2. The range of values is adapted to the respective sensor sizes.
The curves show that the radial-symmetric distortion of the Pyroview 640L compact, due to the larger sensor, takes larger values towards the margin. This displays that the distortion’s determination of larger sensors should not be ignored.
The test field used here was found to be not ideal. However, this was expected from the beginning, as it was chosen to work with less expensive devices. To achieve good and representative numbers, a test field as the one developed by Luhmann [Luhmann et al., 2011] should be used. With this, errors like instability of the object coordinates (due to the aluminum foil, which cannot be fixed to 100%), the shapelessness of the target marks, and a false internal geometry of the coded point marks can be avoided. However, for simple calibrations in order to verify in which range the distortions are, the device used here is sufficient.