Método de detecção de linha do horizonte em imagens marítimas baseado em processamento de imagens

Abstract

Context: The detection of the horizon line is a task that can be used in augmented reality, in port security, in navigation of vessels and autonomous aircraft, among other applications. This line is the edge that forms the boundary between sky and non-sky areas. To find the horizon line can present some challenging elements, such as fog, bad weather, water vapor, reflections and some vessels covering the horizon line. Traditional image processing methods, when used individuality, without the use of other techniques and tools, are not always able to satisfactorily deal with these challenges. On the other hand, methods that use machine learning are getting promising success rates, however, some of these proposed methods involve complex operations, such as deep learning networks, and are executed in GPUs. Since autonomous aircraft have limited hardware and often do not have a GPU, using these methods may not be feasible. Objetive: In this work we propose a horizon line detection method for maritime images using image processing with other methods as support that can be implemented in embedded devices. Method: Using the SMD database, image processing techniques were used, such as edge detection, combined with other methods that do not use machine learning, such as RANSAC, to determine the horizon line in maritime images. Results: the proposed method presented an error in determining the angle below 1 degree for onboard images and between 2.85 and 2.89 degrees for onshore images, and less than 5 pixels of difference from the real line for the estimated line for fifty percent of the tested images. Conclusions: Although the results obtained by the proposed method present an error rate higher than that achieved by the state of the art methods, it can draw an approximate horizon line and can be executed in embedded devices.