Sea Ice Field Analysis Using Machine Vision

Abstract

Sea ice field analysis has motivation in various areas, such as environmental, logistics or ship maintenance. Among other methods, local ice field analysis from ship-based visual observations are currently done by human volunteers and therefore are liable to human errors and subjective interpretations.

The goal of the thesis is to develop and implement a complete process for obtaining dimensions, distribution and concentration of sea-ice floes, which aims at assisting and improving part of the aforementioned visual observations. Such process involves numerous, organized steps which take advantage of techniques from image processing (lens calibration, vignetting removal and orthorectification), robotics (transformation frames) and machine vision (thresholding and texture analysis methods, and morphological operations).

An experimental system setup for collecting the required information is provided as well, which includes a machine vision camera for image acquisition, an IMU device for determining the dynamic attitude of the cameras with respect to the world, two GPS sensors providing a redundant positioning and clock data, and a desktop computer used as the main logging platform for all the collected data.

Through a number of experiments, the proposed system setup and image analysis methods have proved to provide promising results in pack ice and brash ice conditions, thus encouraging further research on the topic. Further improvements should target the accuracy of ice floes detection, and over and under-segmentation of the detected sea-ice floes.