Litcius/Paper detail

Vision-Based AUV Docking to an Underway Dock using Convolutional Neural Networks

Parmeet Singh, Edward Gregson, Jordan Ross, Mae Seto, Chris Kaminski, David Hopkin

202015 citationsDOI

Abstract

A visual positioning scheme is presented as a solution for the final stage (last 15 m) underway docking of the Arctic Explorer AUV (ISE Ltd., Pt. Coquitlam, Canada). Once docked, the AUV can charge its batteries and download its data to the platform that is towing the underway dock. The challenge is that the AUV must be underway as it is towing an array in shallow water. AUV position is detected relative to the underway docking station which is made visible to the AUV camera with a ring of underwater light emitting diodes. The visual positioning scheme is based on a variant of the perspective-n-point algorithm. The mean of the difference between the ground truth and perspective-n-point range is 0.89m with a standard deviation of 0.49m. Similarly, the mean difference for heading and elevation is 1.04 degrees with a standard deviation of 2.09 degrees. These values represent a low error and are acceptable. The next phase is at-sea verification of the undersea dock and visual positioning scheme.

Topics & Concepts

DOCKTowingHeading (navigation)Computer scienceUnderwaterGlobal Positioning SystemGround truthMarine engineeringDocking (animal)Convolutional neural networkArtificial intelligenceComputer visionSimulationEngineeringAerospace engineeringGeologyTelecommunicationsOceanographyNursingMedicineRobotics and Sensor-Based LocalizationUnderwater Vehicles and Communication SystemsWater Quality Monitoring Technologies
Vision-Based AUV Docking to an Underway Dock using Convolutional Neural Networks | Litcius