Automatic Estimation of Ship Size from Satellite Optical Image using Elliptic Characteristics of Ship Periphery

Abstract

As marine trade and traffic volume increase, marine accidents such as collision and sinking between ships are increasing. Remote sensing using satellites is essential to detect missing ships in the event of an accident. We used optical satellite images observing the coast of Korean peninsula to detect ships of various sizes, and to perform accuracy evaluation and error analysis. The maximum likelihood classifier (MLC) technique was used to classify all objects in the image into twelve ships, ship shadows aligned in the northwest direction, linear and v-shaped wakes, and sea. We also applied the ellipse fitting through the classified ship boundary pixels to estimate the length, width, and area of the ship. As aresult, the ship length was estimated to be about 1.21 times and 1,17 times, and the ship width was estimated to be 1.05 times and 1.38 times larger than the actual ship size at the time of satellite image observation. The analysis of the ellipse error was performed based on the eccentricity, the ratio of major axis and minor axis, and the kurtosis of the front of the ship. The correction of the error of the vessel reduced the rms error of the ship length from 7.01 to 2.01. The results of this study show that it is possible to detect the ship missing by optical satellite image in real time and to estimate the accurate ship size in the event of an accident of marine ship.