Multi Modal Mapping of Sea Ice Fields From Remote Shipborne Instrumentation

Abstract

This work investigates the perception bias caused by ice floes occluding open water regions behind them when analyzing sea ice from shipborne cameras. We propose a method to model this effect by predicting an occlusion probability for a given point based on a stochastic representation of the sea ice fields. We further show how the proposed framework can be used to design an unbiased estimator for image-based sea ice concentration by correcting for these occlusions. The occlusion model and proposed sea ice concentration estimators are verified through simulated experiments. The results of this study can be used in algorithms for automated identification and characterization of ice environments from surface ships and are thus an essential step towards improved autonomy in Arctic navigation.