Abstract
In this paper, a method is considered for determining the parameters of a sea vessel, such as its spatial orientation and the distance from the observation point to this vessel from a photographic image to facilitate the task of identifying vessels at night. A measuring system for practical implementation is proposed, consisting of an optical segment, a segment for determining the angle of the vessel by the image, a segment for determining the distance to the object, and a segment for accumulating and filtering data. An algorithm for extracting the parameters of ship lights from a photographic image, their analysis, and the calculation of the quantities required for classification are described. The effectiveness of various classifier architectures for determining the angle of the vessel was experimentally tested, among which the SVM architecture was the most effective. A method for compiling a “depth map” for a static image based on the data of real distances to objects in the daylight image and the coordinates of the corresponding pixels in the same image is described. The method of backpropagation of the error is used for the obtained distances in the corresponding segment of the system based on the existence of the position-distance mapping. The model of the “depth map” constructed based on these data made it possible to obtain a sufficient distance to the object from the photograph.