为了提高视觉惯性里程计在船舶清洗机器人上的定位精度，提出一种融合了双目相机、惯性测量单元和深度传感器的水下定位系统。首先，使用基于灰度世界的白平衡算法与暗通道先验算法对双目图像进行预处理，恢复水下图像真实色彩并进行去雾操作，提高特征提取与匹配精度；然后建立深度传感器数学模型，并将深度数据与视觉关键帧进行对齐，从而在视觉惯性联合优化中加入深度约束，获得高精度位姿估计。试验结果表明，相比于ORB-SLAM3-VIO，本文所提双目-惯性-深度里程计的平均均方根误差减小了41.6%。
To enhance the localization accuracy of visual-inertial odometry on ship cleaning robots, an underwater localization system integrating a stereo camera, an inertial measurement unit, and a depth sensor is proposed. Firstly, the stereo images are preprocessed using a gray-world-based white balance algorithm combined with a dark channel prior algorithm to restore the true colors of underwater images and perform defogging, thereby improving the precision of feature extraction and matching for localization. Then, a mathematical model for the depth sensor is developed, and depth data is aligned with visual keyframes. This allows depth constraints to be integrated into the joint visual-inertial optimization, resulting in high-precision pose estimation. Experimental results conducted in a pool indicate that, compared to the ORB-SLAM3-VIO, the stereo visual-inertial-depth odometry proposed in this paper achieves a 41.6% reduction in average root mean square error.
2025,47(19): 94-98 收稿日期：2024-12-26
DOI：10.3404/j.issn.1672-7649.2025.19.015
分类号：U666; TP242
基金项目：国家重点研发计划资助项目（2020YFC1521704）
作者简介：班慧军(1998-),男,硕士研究生,研究方向为水下定位
参考文献：
[1] HACHICHA S, ZAOUI C, DALLAGI H, et al. Innovative design of an underwater cleaning robot with a two arm manipulator for hull cleaning[J]. Ocean Engineering, 2019, 181: 303-313.
[2] 王继祥, 张增猛, 李洪波, 等. 船体表面海生物水下清洗机器人研究现状及关键技术进展[J]. 船舶工程, 2024, 46(2): 146-164.
[3] 王金科, 左星星, 赵祥瑞, 等. 多源融合SLAM 的现状与挑战[J]. 中国图象图形学报, 2022, 27(2): 368-389.
[4] ZHANG J W, HAN F L, HAN D F. Integration of sonar and visual–inertial systems for SLAM in underwater environments[J]. IEEE Sensors Journal, 2024, 24(10): 16792-16804.
[5] XU Z Z, MARYAM HAROUTUNIAN, ALAN J. An Integrated visual odometry system with stereo camera for unmanned underwater vehicles[J]. IEEE Access, 2022, 10: 71329-71343.
[6] MA S X, WANG J, HUANG Y P, et al. Tightly coupled monocular-inertial-pressure sensor fusion for underwater localization of a biomimetic robotic manta[J]. IEEE Transactions on Instrumentation and Measurement, 2024, 73: 1-11.
[7] 赵洪全, 徐会希. 基于单目直接法的水下视觉惯性里程计方法[J]. 舰船科学技术, 2022, 44(5): 65-69.
ZHAO H Q, XU H X. Underwater visual-inertial odometry based on monocular direct method[J]. Ship Science and Technology, 2022, 44(5): 65-69.
[8] HE K, SUN J, TANG X. Single image haze emoval using dark channel prior[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2010, 33(12): 2341-2353.
[9] HU C, ZHU S Q, LIANG Y M. Visual-pressure fusion for underwater robot localization with online initialization[J]. IEEE Robotics and Automation Letters, 2021, 16(4): 8426-8433.
[10] FORSTER C, CARLONE L, DELLAERT F, et al. On-manifold preintegration for real-time visual-inertial odometry [J]. IEEE Transactions on Robotics, 2017, 33(1): 1-21.
[11] CARLOS C, RICHARD E, JUAN J, et al. ORB-SLAM3: An accurate open-source library for visual, visual-inertial, and multi-map SLAM[J]. IEEE Transactions on Robotics, 2021, 37(6): 1874-1890.
