An Unscented Kalman Filter-based Visual Pose Estimation Method for Underwater Vehicles

Abstract

Pose estimation is a critical technique for underwater vehicles to perform observation and manipulation tasks. Acoustic positioning systems are traditionally used for underwater vehicles, but visual systems have the advantages of low cost, high resolution and high data rates, which have unique advantages in underwater vehicle positioning system. In this paper, an unscented Kalman filter-based visual pose estimation is designed based on the models of underwater vehicles and visual systems. Select the image information of feature points as measurement information, perform the 6DOF pose estimation based on the kinematics model of underwater vehicle. The proposed algorithm is validated by comparative tests with a conventional Perspective-N-Points algorithm, based on the visual data of visual targets collected by a real underwater vehicle.