Evaluation of Camera Calibration Techniques for Quantifying Deterioration

Imaging systems offer an efficient way of obtaining quantitative information on the health status of structural components. They hold particular value for underwater inspections as they can be easily adapted for underwater use and they enable physical information to be captured from a scene for the purpose of later analysis. In order to make the visual data a part of a quantitative assessment, it is necessary to calibrate the imaging systems so that photographed instances of damage can be expressed and measured in physically meaningful real world units, such as millimetres, which can then be used by engineers in subsequent analyses. The imaging system employed in this study is a stereo rig. It consists of two synchronised cameras that capture images of the scene from slightly different perspectives, thereby encoding depth information. This paper evaluates and compares two main approaches for calibrating such a stereo systems, namely, the classical checkerboard procedure and self-calibration based on Kruppa’s equations. Conventional checkerboard calibration must be carried out on-site by photographing a planar checkerboard pattern that is held at multiple random poses, while self-calibration can be carried out after-the-fact and relies only on the static scene acting as a constraint on the camera parameters. The performance of each approach is assessed through a set of experiments performed on controlled real-world specimens as well as on synthetic data. Results indicate that checkerboard calibration is slightly more accurate than self-calibration; however, the practical advantages of using self-calibration may outweigh this reduction in accuracy. An understanding of the advantages and limitations associated with each camera calibration allows inspectors to rationalise the use of either approach as part of their inspection regime, and it helps them to fully capitalise on the benefits of image-based methods.