Ship recognition using optical imagery for harbor surveillance

Many military and homeland defense missions require automated situation awareness in maritime environments. A major element of these missions is automatic detection, tracking, and recognition of ships as they transit harbors. We advocate the use of optical sensors in an Automated Target Recognition (ATR) system to accomplish these missions. This paper reports on the development of maritime optical ATR systems that incorporate a new capability for recognition of known ships, using a database of previously acquired imagery. The approach investigated here uses the local interest point detector and descriptor known as SIFT (Scale Invariant Feature Transform) features. The SIFT interest point detector locates extrema in scale space of Difference-of-Gaussian functions, generating a set of distinctive image regions; the keypoint descriptor measures the orientation of local gradients in the region. The features are normalized, making them invariant to image scaling and rotation and partially invariant to changes in illumination and viewpoint. SIFT features are used in object recognition by matching features extracted from test images with those previously measured in database images. Following feature matching, a geometric verification process is used to eliminate false matches. This paper describes criteria developed to recognize ships using SIFT features and strategies employed to handle changes in camera viewpoint and cluttered backgrounds common in maritime environments.