Visual surveillance has been an active research area due to its crucial role in helping military intelligence and law enforcement agencies to fight against crime and terrorist activities. The goal of a visual surveillance system is to detect abnormal object behaviors and to raise alarms when such behaviors are detected. It is feasible to classify the moving objects in a scene into pre-defined categories, so that their motion behaviors can be appropriately interpreted in the context of their identities and their interactions with the environment.
Detection of small boats in a highly cluttered background environment is an important research activity in port security applications. Recent advances in visible and IR imaging technology improves the ability to observe objects at very long distances, but it is still militarily desirable to identify potential enemy ships/boats, which could be observed at different viewing angles. Several methods have been developed to perform automatic detection and classification of big ships (e.g. aircraft-carriers, combat ships, transportation ships, etc) from images, but none of them is suitable for small boat classification. The small boat regions in grayscale IR images can initially be segmented using the Graph-cut algorithm and those in visible images by employing a novel Adaptive Progressive Thresholding (APT) algorithm. A Principal Component Analysis (PCA) based multi-view small boat classification system is being developed for visible and IR small boat images which are captured in a highly cluttered background environment.