Automatic texture and anomaly mapping in under-ice video datasets

Abstract

The exploration of under-ice environments has seen increased interest over the past few years due to advances in technological capabilities, such as autonomous underwater vehicles (AUVs), as well as interest in exploration of polar regions and Jupiter's ice-covered moon Europa. Searching for interesting features under the ice, including animals capable of sustaining life in such harsh environments, is of great interest in both polar (Antarctica) and planetary (Europa) domains. Underice environments, such as those encountered beneath the Antarctic ice shelves, are largely devoid of such features and tend to be monochromatic centered on the blues of the ice. Postprocessing of under-ice datasets can be very tedious for human analysts. Presented here are algorithms to aid in the postprocessing of such large and mostly featureless datasets. Two novel algorithms are presented here which use point-feature detections in video frames to estimate texture (number and spread of features) and anomaly locations (dense groupings of features). Two additional algorithms are proposed which use hue-based methods to estimate the percentage of non-ice pixels present in the video frames and to detect anomalous colored pixel groups corresponding to candidate anomalies against the background of the ice. These algorithms are presented herein along with results from testing with both simulated and realworld under-ice video datasets.