Automatic image segmentation for boundary detection of apparently connected sea-ice floes

Abstract

Ice concentration, ice types, and floe size distribution are three important factors in the analysis of ice-structure interaction in a broken ice field. The use of cameras as sensors for offshore operations in ice-covered regions is relevant for identification of the ice conditions and ice parameters, as part of a sea-ice monitoring system. In particular, this can be used for estimation of ice forces that are critical to Dynamic Positioning (DP) operations in Arctic waters. For the actual ice image, the apparent connection between ice floes should be identified, as it challenges the boundary detection algorithms and seriously affects ice floe size analysis. To solve this problem, a method of connected ice floes segmentation based on the watershed transform is presented in this paper. After ice image preprocessing, all the ice floes in the image are first identified by the Otsu threshold algorithm, and the greyscale image is converted into a binary image. Then, the seed points for the individual ice floes are located, and the watershed algorithm is carried out to segment the image. Finally, a chain code is used for concave spot localization, and then over-segmentation lines are removed. Experimental results are presented and a discussion of the method is given.