A featureless approach to improve self-consistency in structured light bathymetry

Abstract

This paper describes a novel method for calibrating structured light setups using a featureless approach to quantify and improve self-consistency in bathymetric maps. The self-consistency and accuracy of seafloor reconstructions and information derived from them are limited by uncertainties in vehicle localisation, sensor models and their calibration. For high-resolution setups such as structured light, these uncertainties can be several orders of magnitude larger than the resolution of the maps generated. Although techniques such as simultaneous localisation and mapping and bundle adjustment can correct pose estimates and sensor calibrations to improve map self-consistency, both methods typically rely on finding and matching features in the data, which limits their application to structured light since a key advantage of this method is that it does not rely on seafloor features to be present in order to work. In this paper, we develop a fully featureless approach to improve self-consistency in structured light setups. Simulations are performed to validate the proposed method, and we analyse data that was collected using the Autosub6000 autonomous underwater vehicle equipped with the BioCam seafloor mapping instrument at a depth of approximately 1000m in the Darwin Mounds UK marine protected area. The results for independent parameter optimisation demonstrate that the fully featureless approach can converge towards optimal calibrations.