DOI: 10.14489/vkit.2017.10.pp.003-012

Тихомирова Т. А.
МЕТОД ВЫБОРА ВИЗУАЛЬНЫХ ОРИЕНТИРОВ ДЛЯ ЗАДАЧИ НАВЕДЕНИЯ
(c. 3-12)

Аннотация. Предложен метод выбора малоразмерных визуальных ориентиров (реперов) на контурном изображении ландшафта. Сформулированы критерии отбора реперов по признакам яркости, формы, изолированности и связи с протяженными объектами фона. Для оценки критериев применен дескриптор, вычисляемый с помощью быстрого преобразования Хафа и описывающий яркость / направление альтернативных касательных к контурам, проходящим через локус. Представлена сравнительная оценка эффективности различных методов оконтуривания в контексте рассматриваемой задачи.

Ключевые слова:  анализ изображений; визуальный ориентир; контурный анализ; преобразование Хафа; касательное накрытие.

Tikhomirova T. A.
VISUAL LANDMARKS SELECTION FOR THE AIMING TASK
(pp. 3-12)

Abstract. A method for selecting visual landmarks for the aiming task is presented. Landmarks are selected on a contour image of a landscape  according to a number of criteria – contrast (or brightness), form, uniqueness and connection with the global background object. To evaluate the criteria, we use a special data structure – the Field of Alternative Tangents (FAT), which is calculated by means of a fast local Hough transformation. The FAT contains the directions and the average brightness of eight tangent segments corresponded to eight most probable directions of the contours passing through the locus. The criterion of contrast is defined as the brightness of the main tangent, called the weathervane. When tracking the bench, the weathervane is used to track the course of the aircraft. The criterion of form – flag – is estimated as the product of the brightness of the other tangent segment called shaft to the sine of it’s angle to the weathervane. Shaft is selected from FAT and maximizes the flag. The directions of the weathervane and the shaft are the main characteristics of the landmark geometry. The criterion for the landmark uniqueness is defined as the distance to it’s nearest neighbor selected by the same criteria. Landmarks are linking to global terrain objects by fast tracing of the contour image by short polygons consisting of tangent segments presented by FAT. When searching for landmarks, a rank strategy is used: for each criterion, a fixed number of landmarks is selected successively, and the criterion value for the worst of them becomes a feature of the input image in the context of this criterion. Such an approach guarantees the obtaining of the required number of landmarks and allows estimation of the informativeness of the image. The results of comparing the informativeness of contour images obtained using various edgedetecting algorithms are presented. Integro-differential filters of boundaries and lines based on the local Hough transformation are proposed and their advantages are demonstrated.

Keywords: Image processing; Visual landmarks; Contour analysis; Hough transform; Tangent cover; Image processing; Visual landmarks; Contour analysis; Hough transform; Tangent cover.