Identifying Sources of Error in Underwater Photogrammetry via Sensitivity Analysis

In aiming for sub-millimeter 3D measurement accuracy for underwater objects such as shellfish in multi-view imagery, we investigate the effect of errors in camera extrinsic and intrinsic parameters, pixel feature matching, and optical system properties. Multi-view pixel ray intersection (triangulation) is used to calculate 3D object feature positions and by extension object dimensionality. By perturbing dependant variables we conduct a numerical sensitivity analysis for the error in dimensional measurement, and discover in order to achieve a measurement accuracy of less than 1mm, the pixel features must be localised to within 5 pixels, and error in camera translation and rotation must be no greater than 5mm and 0.5 degrees respectively. We also investigate environmental and systematic sources of error such as refraction through a thermocline, and find only minimal effects on the measurement error.