基于被动式三目视觉的核电厂水下定位系统及方法研究

Research on Global Underwater Positioning System and Method for Nuclear Power Plants Based on Passive Trinocular Vision

  • 摘要: 针对核电厂水下约束空间中难以鲁棒、连续获取全局定位信息的问题,设计了一种基于三目视觉和被动式靶标相结合的全局定位系统。结合三目相机组与两轴高精度转台研制了一种水下定位装置,提出了覆盖全局坐标系建立、相机内外参数联合优化、相机组与两轴转台偏心标定的全参数标定方法,构建了基于多视图几何的实时三维重建解算与逆向全局定位变换框架,实现了大范围转台动态跟踪下目标全局坐标的统一输出。在封闭式模拟水池中,分别在1、3、5 m距离下开展了静态绝对定位、刚体相对定位及动态跟踪性能试验。试验结果表明:5 m水深条件下,静态单靶标绝对定位平均误差为0.871 mm;在1~5 m测距范围内,静态刚体相对定位最大误差为1.245 mm;5 m水深大范围动态跟踪工况下,相对定位平均绝对误差为11.959 mm。该系统部署便捷,可在动态大范围工况下实现高精度定位,能够有效满足核电厂复杂水下环境的作业需求,可为核电水下定位相关技术开发与装备研制提供技术参考。

     

    Abstract: Acquiring robust and continuous global positioning information for underwater targets in nuclear power plants remains a critical engineering challenge due to confined operational spaces, complex water refraction, and severe multi-path reflections from stainless-steel pool walls. To address these limitations, this paper proposed a novel underwater global positioning system that intelligently integrated a passive trinocular vision architecture with a high-precision two-axis tracking turntable. Rather than relying on traditional localized visual odometry, the proposed system was explicitly designed to establish a unified and highly accurate global coordinate framework for large-scale spatial tracking. Initially, a specialized underwater optomechanical device was developed, utilizing blue-light active illumination and narrow-band filtering to effectively suppress ambient optical noise. Methodologically, a comprehensive full-parameter calibration pipeline was introduced. This pipeline seamlessly encompassed the establishment of a static global coordinate system, the joint optimization of camera intrinsic and extrinsic parameters to compensate for underwater refraction, and the spatial eccentricity calibration between the camera array and the turntable’s mechanical rotational axes via algebraic spherical fitting. Furthermore, a real-time 3D reconstruction algorithm based on multi-view epipolar geometry and an inverse kinematic spatial transformation model were constructed. This algorithmic framework enables the continuous and unified output of the target’s absolute global coordinates, even during large-scale dynamic tracking. To rigorously validate the system’s overall performance, a 5-meter-deep simulated closed stainless-steel pool test platform was constructed. Extensive experiments were systematically conducted at varying radial distances of 1, 3, and 5 m. Quantitative results demonstrate exceptional accuracy: The average absolute positioning error for a single target is 0.871 mm during a 100-500 mm reciprocating motion; The maximum relative positioning error for a static rigid body is 1.245 mm within a 5 m radius. Crucially, during challenging large-range dynamic tracking at a 5 m depth, the system maintains a highly reliable average absolute error of 11.959 mm. Ultimately, the proposed system exhibits high precision, strong anti-interference capabilities, and excellent dynamic adaptability, providing a highly deployable spatial localization solution for complex underwater maintenance and inspection tasks in the nuclear power industry.

     

/

返回文章
返回