Target Analysis for the Multispectral Geometric Calibration of Cameras in Visual and Infrared Spectral Range

Iterative feature detection of a coded checkerboard target for the geometric calibration of infrared cameras

“…The cost function of the nonlinear optimization is based on minimizing the total Euclidean distance (RMSE, reprojection error) between the given points and their reprojection. Schramm et al (2021) show that the uncertainty in the calibration depends on the uncertainty in the reference data, i.e., the quality of the feature localization.…”

“…These targets often achieve very good results in the long-wavelength infrared (LWIR) and mid-wavelength infrared (MWIR) due to their high contrast compared to emissivity-based targets, but they require a complex setup and are thus more expensive and heavier. Also, the results of extrinsic calibrations with VIS cameras are worse (Schramm et al, 2021). Thus, the question arises of how to improve the calibration quality of classical emissivity-based targets compared to the state of the art.…”

“…The sizes of the target and the features have a crucial impact on the detection. The optimal checkerboard target size in relation to the camera's FOV is discussed in Schramm et al (2021). The size of the centered ArUco marker is also a trade-off.…”

“…For better comparability with the uncoded checkerboard target from Schramm et al (2021), the same target dimensions (600 mm × 450 mm) and checkerboard side lengths (50 mm) were used in the present work. This meant that a white border and full-length checkerboard squares at the target border were not needed for detection, allowing another line of checkerboard corners to be used at the outermost edge.…”

“…In this section, the experimental results from the coded target presented above and the uncoded checkerboard target from Schramm et al (2021) are compared. In our experiments, an LWIR camera (Optris PI 450) was used, which is sensitive to radiation in the spectral range from 7.5 to 13 µm.…”