Detection and localization of underwater targets based on monocular vision

Abstract

Monocular vision tracking, binocular vision tracking and multi-purpose visual tracking are the main current computer vision positioning tracking patterns. Due to the small field of view, the complex system structure and the difficulty of calibration matching, the binocular vision tracking system is being replaced by the monocular vision tracking system in some applications. The monocular tracking system has simple structure, and less calibrations. This paper investigates the recognition and localization of underwater moving targets using monocular vision. The color threshold algorithm is used to detect and extract the target color information. After this, a combination of geometric model method and data fitting method is proposed to detect the moving target and to obtain it's relative distance to underwater camera. Finally, we established an underwater experiment system with a template target, an underwater camera, a lighting system, and a PC computer. The experimental verification in the laboratory demonstrates the good performance of our proposed detecting and locating method.