An Efficient Lightweight 2D Driven 3D Detector for Underwater Robots Tracking

Abstract

We focus on the research of detection, localization and tracking of an underwater remotely operated vehicle (ROV) in underwater environment. Previous related work focused on using either only 2D images or only 3D point clouds for detecting and tracking underwater objects, resulted in lacking of either 3D position information or low robustness to challenging underwater environment such as strong occlusion. Instead, we make use of the advantages of mature 2D detector and advanced 3D point clouds processing technique to build a complete 2D driven 3D detector and tracker pipeline. The proposed framework allows to detect and track the 3D pose of a target ROV at short distance, with high accuracy and high robustness. Our method was evaluated in underwater environment by dynamic tracking a real target ROV in leader-follower formation experiment. The experimental results demonstrated the effectiveness and efficiency of our proposed method.