Low cost and low power multispectral thermal-visible calibration
Axel Beauvisage, Nabil Aouf · 2017
Because it is a prior step for multiple computer vision applications, camera calibration is a critical task and needs to be performed accurately. With the emergence of new multispectral systems, the need of simple and reliable crossmodality calibration techniques has arisen. In this paper, we present a new calibration process for visible/far infrared(IR) stereo setups. Our calibration board is made of wood, on which high power Light Emitting Diodes (LED) have been placed. The pattern formed can thus be seen in both modalities without the need to heat it. We used a limited number of LEDs to keep the power consumption to a low level. The proposed solution is therefore cheap, easy to manufacture and does not require a lot of energy, which makes it very portable and convenient to use. We prove that the reduced number of calibration points does not affect the accuracy of the calibration process, as a mean reprojection error less than 0.3 pixels is obtained with both cameras. We also show the suitability of the calibration for 3D reconstruction by computing the 3D position of the board with our calibrated stereo setup.