AUTOMATED TRACKING OF ICE FLOES - A STOCHASTIC APPROACH

A great deal of effort has been expended on developing and implementing algorithms for tracking sea ice in satellite images. Most of the algorithms involve extracting a set of features from each of a pair of images, defining a suitable measure of similarity and then matching pairs of features using the similarity measure. These algorithms relegate the spatial information, inherent in the location of the features within the image, to the implementation stage where it is used to limit the search area. They also have limited capabilities for assessing the quality of the final result. This paper presents an approach, based on probability distributions, that directly incorporates the spatial information about feature locations into the estimation procedure. The basic idea is, given an image taken at time t0, use a probability model to determine how features in the image will appear at time t1, then use the probability distribution to identify features common to both images. The use of a probability model provides a means to measure the goodness of fit of the resulting matches. The features that will be used in this paper are the outlines of sea ice floes observed in SAR images, although the method can be applied to any set of features. The floe outlines are found using an erosion-propagation algorithm which combines erosion from mathematical morphology with local propagation of information about floe edges.