Identification and tracking of marine objects for collision risk estimation.

Abstract

With the advent of modem high-speed passenger ferries and the general increase in maritime traffic, both commercial and recreational, marine safety is becoming an increasingly important issue. From lightweight catamarans and fishing trawlers to container ships and cruise liners one question remains the same. Is anything in the way? This question is addressed in this thesis. Through the use of image processing techniques applied to video sequences of maritime scenes the images are segmented into two regions, sea and object. This is achieved using statistical measures taken from the histogram data of the images. Each segmented object has a feature vector built containing information including its size and previous centroid positions. The feature vectors are used to track the identified objects across many frames. With information recorded about an object's previous motion its future motion is predicted using a least squares method. Finally a high-level rule-based algorithm is applied in order to estimate the collision risk posed by each object present in the image. The result is an image with the objects identified by the placing of a white box around them. The predicted motion is shown and the estimated collision risk posed by that object is displayed. The algorithms developed in this work have been evaluated using two previously unseen maritime image sequences. These show that the algorithms developed here can be used to estimate the collision risk posed by maritime objects.