6D Pose Estimation for Subsea Intervention in Turbid Waters

Manipulation tasks on subsea instalments require extremely precise detection and localization of objects interest. This problem is referred to as “pose estimation”. In this work, we present a framework for detecting predicting 6DoF pose relevant (fish-tail, gauges, valves) panel under varying water turbidity. A deep learning model that takes 3D vision data an input developed, providing more robust 6D estimate. Compared the 2D model, proposed method reduces rotation translation prediction error by (−Δ0.39∘) (−Δ6.5 mm), respectively, in high turbid waters. The approach able provide object well estimation with average precision 91%. results show 2.59∘ 6.49 cm total deviation compared ground truth unseen turbidity levels. Furthermore, our runs at over 16 frames per second does not refinement steps. Finally, facilitate training such also collected automatically annotated new underwater dataset spanning seven levels