水工隧洞环境下基于视觉增强-惯性SLAM的ROV水下定位技术研究

Research on ROV underwater positioning technology based on enhanced visual-inertial SLAM in hydraulic tunnel environments

摘要

【目的】随着我国引调水工程数量的逐渐增多，构建高效率的智慧巡检体系对引调水工程的长期安全运维具有重要意义。水工隧洞内的精确定位是ROV实现自主化巡检的关键和难题，而现有ROV水下定位方法主要面向海洋环境，常规定位方法在水工隧洞内难以发挥作用。【方法】提出一种基于视觉增强-惯性SLAM的水下定位技术。针对隧洞图像特征区域受限且特征模糊等问题，构建了基于DE-MFET的水下图像增强网络，融合深度信息和多尺度特征强化模块增强图像重要区域特征的通道响应，突出水下图像边缘局部细节特征，提高SLAM关键帧之间的特征匹配数量和效率；针对水工隧洞内特征弱及特征一致性强等导致视觉里程计易失效的问题，将增强的视觉里程计与惯性导航数据进行融合，进一步校正ROV瞬时位姿，提高ROV在水工隧洞中的定位精度。【结果】通过自制的水工隧洞数据集、LSUI和UIEB公开水下数据集验证提出的图像增强方法使水下图像质量显著提高，在UCIQE、UIQM评分值和ORB特征匹配数量方面均优于其他图像增强方法。并在湖北省鄂北水资源配置工程中进行ROV水下定位试验，结果表明视觉增强-惯性SLAM定位方法有效提升了ROV在水工隧洞内的定位精度。【结论】该定位方法在水工隧洞环境下具有一定的优越性和鲁棒性，为引调水工程智慧巡检方案提供重要研究基础。

Abstract

[Objective] With the increasing number of water transfer projects in China, establishing an efficient intelligent inspection system is crucial for the long-term safe operation and maintenance of water transfer projects. Precise positioning within hydraulic tunnels is key and challenging for ROVs to achieve autonomous inspection, while existing ROV underwater positioning method primarily target marine environments, making conventional positioning approaches difficult to apply effectively in hydraulic tunnels.[Methods] An underwater positioning technology based on enhanced visual-inertial SLAM is proposed. Aiming at the problems of limited feature regions and blurred features in tunnel images, an underwater image enhancement network based on DE-MFET is constructed. This network fuses depth information and multi-scale feature enhancement modules to enhance the channel response of important image features, highlight the local edge details of underwater images, and improve the quantity and efficiency of feature matching between SLAM keyframes. To counteract frequent visual odometry failures caused by weak features and strong feature homogeneity in hydraulic tunnels, enhanced visual odometry is fused with inertial navigation data to further correct the ROVs instantaneous pose, improving positioning accuracy in hydraulic tunnels.[Results] Validation using a self-developed hydraulic tunnel dataset, the LSUI and the UIEB public underwater dataset demonstrates that the proposed image enhancement method significantly improves underwater image quality, outperforming other enhancement method in UCIQE, UIQM scores, and ORB feature matching quantity. ROV underwater positioning experiments in Hubei's Ebei Water Resources Allocation Project confirm that the enhanced visual-inertial SLAM method effectively improves positioning accuracy within hydraulic tunnels.[Conclusion] This positioning method exhibits notable superiority and robustness in hydraulic tunnel environments, providing a critical research foundation for intelligent inspection solutions in water transfer projects.

关键词

视觉-惯性SLAM / 水下图像增强 / 水工隧洞 / ROV水下定位 / 引调水工程 / 智慧巡检 / 信息融合 / 深度估计网络

Key words

vision-inertial SLAM / underwater image enhancement / hydraulic tunnel / ROV underwater positioning / water diversion project / intelligent patrol inspection / information fusion / depth estimation transformer

基金资助

国家自然科学基金项目(42302337)

湖北省智慧水电技术创新中心2023 年度开放研究基金资助项目(SDCXZX-JJ-2023-09)

湖北省教育厅科学研究计划重点项目(D20231304)

深地国家科技重大专项(2024ZD1001003)