Water-MBSL: Underwater Movable Binocular Structured Light-Based High-Precision Dense Reconstruction Framework

Structured light systems are widely used in underwater dense reconstruction due to their excellent accuracy. However, the current related methods mainly focus on fixed positions. The reconstruction performance in motion is insufficient. Therefore, we propose an underwater movable binocular structured light (MBSL) based high-precision dense reconstruction framework, named WaterMBSL, to realize the robot reconstruction while moving. Specifically, an onboard binocular structured light system based on mirror-galvanometer is developed first. Then, a simplified underwater point cloud acquisition algorithm is presented to quickly obtain 3-D information of the scene. Besides, a new underwater motion compensation algorithm combining inertial measurement unit and uniform velocity model is proposed. Moreover, the generalized-ICP point cloud registration algorithm is introduced to achieve accurate motion estimation. Finally, an underwater movable reconstruction platform is developed by integrating the self-designed structured light system with the underwater robot BlueROV for validating the performance of our proposed Water-MBSL. Experimental results show that satisfactory motion reconstruction performance can be obtained.

I. Introduction

In Recent years, underwater robots have been widely used in underwater operations, such as underwater pipeline monitoring [1], defect detection [2], inspection [3], and archaeology [4]. By obtaining 3-D information of underwater scenes, robots can better perform these tasks. Therefore, underwater 3-D reconstruction has important research significance. As a common reconstruction method, vision is mainly divided into two forms: passive vision and active vision. Although there have been some studies based on passive vision to achieve robot navigation [5] and localization [6], the lack of light and the scarcity of scene texture features make passive vision relying on scene image feature matching ineffective for underwater reconstruction. Fortunately, active vision 3-D reconstruction methods represented by structured light systems [7], [8] are highly adaptable to the environment. The line structured light system (LSLS) has been shown to be more suitable for underwater environments [9], [10]. Depending on the scanning method, LSLS can be divided into three categories: single-laser LSLS (SLLSLS) [11], [12], motor-driven LSLS [13], [14], [15], and self-scanning LSLS (SSLSLS) [16], [17].