Abstract

This paper presents a system for detecting horizontal features in nautical scenes. Horizontal features can be used to reduce the search space in stereo vision correspondence. The system is a part of ongoing research into maritime craft tracking. Detection of these features is based on an algorithm that employs a modified template matching method. The template matching is reduced to regions obtained using variable window segmentation. The parameters of the template used in matching are dependent on the horizontal position in the image. This dependency accounts for the perspectiveness in the scene arid is automatically determined from the external camera calibration parameters. It brings an improvement in localisation of the detected horizontal features in the image. The algorithm provides a robust feature detection and a significant reduction in the processing resources for stereo matching algorithm by reducing the search area. The horizontal features can also be used for estimation of the range using a single camera where the camera calibration parameters are known.