The Computer Vision System for a Crewless Boat

The article considers a computer vision system for a crewless ship. The existing approaches to the construction of such systems are described, and the strengths and weaknesses of different methods are analyzed. The hardware part of the system is considered. The functional scheme of the system and technical characteristics of the equipment are given. An approach to solving the problem of detecting a given class of objects based on the YOLOv5 algorithm is described. The accuracy of the obtained model is confirmed with the help of precision-recall curve. The paper also presents an algorithm for calculating the heading angle for the dynamic positioning of an uncrewed vessel based on data from the vision system. The advantage of the proposed approach is its fast performance, which allows the use of onboard computers in real-time. The highly effective strategy allows the onboard system's image to be broadcast over a local network without losing image quality. The performance of the computer vision system and the algorithm for detecting a given class of objects is verified using long-term hard-loaded testing. The test results showed good accuracy and stability of the developed system.