Embedded vision systems for ship recognition

Abstract

Maritime security includes reliable identification of ship entering and leaving a nation's territorial waters. Automated systems that could identify a ship could complement existing electronic signal identification methods. The use of Forward Looking Infrared (FLIR) and Synthetic Aperture Radar (SAR) enables ship image acquisition round-the-clock but their cost and complexity means few installations are available. The use of lower cost embedded vision systems using visible light for surveillance in a low-bandwidth sensor network could complement existing surveillance methods to improve surveillance coverage. This paper presents an overview of automatic ship detection methods with a view towards embedded implementation of suitable algorithms on optical smart cameras. We present results on applying Hu's moment invariants for feature extraction on several classification algorithms. We achieved accuracies of close to 80% using the KStar and multilayer perceptron classifiers in recognizing one of four ship classes.