UNDERWATER NON-LINEAR CAMERA CALIBRATION: AN ACCURACY ANALYSIS

Abstract

One of the most challenging problems associated with underwater 3D movement analysis is the accurate calibration of the cameras. Additional sources of errors are present in underwater acquisitions such as the nonlinear distortion caused by water interface, camera lenses (ex. wide angle) and housing’s glasses. Despite this, in the literature, systems based on a linear calibration model (DLT) were proposed (Yanai et al., 1996; Machtsiras & Sanders, 2009; Gourgoulis, et al. 2008). However, the results of underwater accuracy were not similar to those obtained out of the water. In Kwon, et al. 1999, the use of a modified DLT algorithm to model the distortion was proposed but the results of accuracy were not substantially improved, with Root Mean Square (RMS) values ranging from 5.6 to 7.2mm. Recently, alternative approaches were proposed to non-linear camera calibration and submillimeter accuracy was reached (Cerveri et al., 1998; Zhang, 2000; Pribanić, Sturm & Cifrek, 2008). However, these approaches were not applied underwater. In previous work, a new non-linear calibration method using a straight line plane object was proposed and tested out of the water (Silvatti et al., 2009 available in http://calib.googlecode.com). In this work, this novel method was tested in underwater conditions and its accuracy evaluated.