Identification and tracking of maritime objects in near-infrared image sequences for collision avoidance

Abstract

This paper describes the continuing development of an image processing system for use on high-speed passenger ferries. The system automatically identifies objects in a maritime scene and uses the detected motion to alert a human observer to potential collision situations. Three integrated image-processing algorithms, namely an image preprocessor, a motion cue generator, and a target tracker, perform the identification and tracking of maritime objects. The pre-processing filters the image and applies a histogram technique to segment the sea from potential objects of interest. The segmented image is passed to the motion cue generator, which provides motion cues based on the differences between consecutive frames of segmented image data. The target tracker applies dynamic constraints on object motion to solve the correspondence problem, thus increasing the confidence that an identified object is a target. Identified and tracked objects are highlighted to a human observer using a white box viewing cue placed directly around the object of interest.