Tightly Coupled Binocular Vision-DVL Fusion Positioning Feedback for Real-time Autonomous Sea Organism Capture

Abstract

Visual positioning is very significant for underwater robot autonomous manipulation. However, visual positioning feedbacks are often ineffective with unstable and discontinuous matching information during small target capture in underwater environment. Therefore, the low cost sea organism object positioning process should be applied with self-navigation and matching system in order to guide the underwater vehicle manipulator systems (UVMS) manipulation control. In order to improve the matching robustness and accuracy, this study established the underwater refraction and intrinsic & extrinsic parameter optimization model accurate calibration, designed an improved ORB feature matching method, with adaptive threshold to extract FAST feature points and non-maximum suppress to remove feature point blocks in the feature extraction stage. Moreover, the comparison of three pixel blocks norm values are proposed in the feature point description stage to improve the robustness of the descriptor. This paper has proposed a novel tightly coupled binocular vision-DVL fusion positioning feedback optimization algorithm with multi-state constraint Invariant Extended Kalman Filter to improve accurate real-time measurement for the capture of sea organism with limited features. Systematic tank and oceanic experiments have been performed to prove and analyze the purposed algorithm. Experimental results show that the algorithm can provide reliable positioning information in tank and complex seabed environment, which is effective for the underwater robot to realize accurate capture of marine organisms.