Real-Time Perception and Positioning for Creature Picking of an Underwater Vehicle
Shaowei Cui, Yu Wang, Shuo Wang, Rui Wang, Wei Wang, Min Tan
Abstract
This study addresses the ability of an underwater vehicle-manipulator system (UVMS) to perceive submarine creatures in the complex undersea environment. A real-time system for Submarine Creature Perception and Positioning (SCPP) based on a binocular vision system is proposed. In particular, we adopt a new Four-Corner Neighborhood Matching Method (FCNMM) to obtain specific target positions. A Fusion Correction Mechanism (FCM) is added to the Kernelized correlation filter (KCF) tracking algorithm to improve tracking performance. Furthermore, SCPP is applied in our UVMS platform to perform seafloor creature picking at the Zhangzidao seafood farm in Dalian, China. The experimental results show the feasibility and robustness of the SCPP system.