Underwater Circular Object Positioning System Based on Monocular Vision

Abstract

Underwater target positioning technology is a key technology in the field of marine engineering and has broad application prospects. Compared with binocular positioning, monocular positioning has the characteristics of simple equipment, high real-time performance, and no matching difficulty. However, due to the complex underwater environment and the lack of depth information, monocular positioning still has many difficulties applied to underwater. In this paper, an automatic recognition and three-dimensional positioning system for circular objects with known dimensions based on monocular vision is designed. Firstly, a circular target feature point automatic extraction technique is designed to obtain the pixel coordinates of four feature points. Then, according to whether the focus is on the interface between water and air, the underwater refraction process is modeled in two cases. We focus on the relationship between underwater imaging points and air imaging points, and finally achieve underwater camera calibration. Finally, combining the world coordinates of its four feature points, the EPnP method is selected to finally determine the target three-dimensional information. The experimental results demonstrate that the proposed method can quickly identify circular objects and get the position of the target without other depth sensors, which can meet the requirements of underwater robots well.