Development of 3D Visualisation Algorithms for the Effective Interpretation of Tunnel Subsurface Radar Data

开发 3D 可视化算法以有效解释隧道地下雷达数据

基本信息

批准号：
2440423
负责人：
金额：
--
依托单位：
Newcastle University
依托单位国家：
英国
项目类别：
Studentship
财政年份：
2020
资助国家：
英国
起止时间：
2020 至无数据
项目状态：
已结题

来源：
https://gtr.ukri.org/projects?ref=studentship-2440423
关键词：
Development 3D Visualisation Algorithms Effective

项目摘要

Relevant EPSRC Research Areas:Architectures and operating systems; Artificial intelligence technologies; Digital signal processing; Graphics and visualisation; Ground Engineering; Human-communication interaction; Image and vision computing; Infrastructure and urban systems; Mathematical physics; Numerical Analysis; Operational Research; Programming languages and compilers; Software Engineering; Structural Engineering Project Synopsis:Anticipated growth of global rail network demand is set to further increase pressure on aging tunnel infrastructure. The inherent safety risks, severity of structural compromise and ever reducing allocatable resources for targeted maintenance, has prompted development of TRACKSCAN, an infrastructure inspection hardware solution based on Ground-Penetrating Radar (GPR), able to rapidly produce comprehensive 360-degree point-cloud tunnel structure profiles, referred to here on as 360-GPR data. However, this hardware currently lacks a software counterpart that can clearly convey this data to human operatives. This project proposes the development of bespoke visualisation software to translate raw 360-GPR data from TRACKSCAN into immersive, intuitive, 3D digital survey environments to facilitate non-disruptive off-site inspections of tunnel structural integrity through the use of Virtual Reality (VR) hardware. The main challenge to overcome is that the GPR data concerned is inherently non-planar, but helical, as TRACKSCAN utilises a unique rotary kinematic antenna array. Subsequently, leading algorithms for GPR data processing and visualisation, developed across numerous independently conducted research projects, cannot function on this data. Ergo, the primary research goal of this project is the formulation, implementation, and unification of novel procedural augmentations to these strategies for the development of algorithms optimised to enrich, segment and triangulate subsurface geometry from pioneering 360-GPR data. The secondary challenge of this project is to address the prominent lack of interpretive clarity associated with feature identification in conventional GPR data displays, which has made structural information obtained inaccessible to non-radar trained rail industry operatives. Therefore, visualiser development will introduce a deep learning-based object classification scheme to dynamically highlight and qualitatively assess subsurface features (i.e. assets, defects, and medium characteristics) present within 360-GPR data in real time. Extracted information will then be intuitively conveyed to the end-user by design and integration of a dedicated streamlined user interface. Additionally, this project seeks to improve achievable levels of operative immersion within the survey data, facilitating more ergonomic execution of virtual inspections and permitting unrestricted analysis of structural features onsite teams cannot easily view (e.g. narrow culvert interiors). This will be achieved through the creation of an original rendering engine, capable of translating subsurface geometry interpolated from the 360-GPR data into realistically textured 3D virtual environments. Navigation will utilise a commercial VR hardware unit, placing operatives directly within the digitally reconstructed tunnels, whilst texture realism will stem from the development of new algorithms to fuse 360-GPR data with corresponding near-surface 3D Lidar scans.Research proposed will therefore centre on the Development of 3D Visualisation Algorithms for the Effective Interpretation of Tunnel Subsurface Radar Data generated by the new Infrastructure Inspection Radar system TRACKSCAN. This work will involve software engineering; practical testing using VR hardware; topics in machine learning, computational geometry, and human computer interaction within the gaming industry; alongside investigation of the physics and mathematics behind 360-GPR data inversion.

EPSRC相关研究领域：架构和操作系统；人工智能技术；数字信号处理；图形和可视化；地面工程；人机交互；图像与视觉计算；基础设施和城市系统；数学物理；数值分析；运筹学；编程语言和编译器；软件工程；结构工程项目简介：全球铁路网络需求的预期增长将进一步增加老化隧道基础设施的压力。固有的安全风险、结构受损的严重性以及用于有针对性的维护的可分配资源的不断减少，促使 TRACKSCAN 的开发，这是一种基于探地雷达 (GPR) 的基础设施检查硬件解决方案，能够快速生成全面的 360 度点云隧道结构剖面，此处称为 360-GPR 数据。然而，该硬件目前缺乏可以将这些数据清楚地传达给人类操作员的软件对应物。该项目建议开发定制可视化软件，将 TRACKSCAN 的原始 360-GPR 数据转换为沉浸式、直观的 3D 数字测量环境，以便通过使用虚拟现实 (VR) 硬件对隧道结构完整性进行无中断的场外检查。需要克服的主要挑战是，所涉及的探地雷达数据本质上是非平面的，而是螺旋形的，因为 TRACKSCAN 采用了独特的旋转运动天线阵列。随后，在众多独立开展的研究项目中开发的探地雷达数据处理和可视化的领先算法无法在这些数据上发挥作用。因此，该项目的主要研究目标是制定、实施和统一这些策略的新颖程序增强，以开发优化的算法，以从开创性的 360-GPR 数据中丰富、分割和三角测量地下几何形状。该项目的第二个挑战是解决传统探地雷达数据显示中与特征识别相关的明显缺乏解释清晰度的问题，这使得未受过雷达培训的铁路行业操作人员无法获得结构信息。因此，可视化工具的开发将引入基于深度学习的对象分类方案，以实时动态突出显示和定性评估 360-GPR 数据中存在的地下特征（即资产、缺陷和介质特征）。然后，通过设计和集成专用的简化用户界面，将提取的信息直观地传达给最终用户。此外，该项目旨在提高测量数据中可实现的操作沉浸水平，促进更符合人体工程学的虚拟检查执行，并允许对现场团队无法轻松查看的结构特征（例如狭窄的涵洞内部）进行不受限制的分析。这将通过创建原始渲染引擎来实现，该引擎能够将从 360-GPR 数据插值的地下几何图形转换为逼真纹理的 3D 虚拟环境。导航将利用商用 VR 硬件单元，将操作员直接置于数字重建的隧道内，而纹理真实感将源于新算法的开发，以将 360-GPR 数据与相应的近地表 3D 激光雷达扫描融合。因此，拟议的研究将集中于开发 3D 可视化算法，以有效解释由新的基础设施检查雷达系统 TRACKSCAN 生成的隧道地下雷达数据。这项工作将涉及软件工程；使用 VR 硬件进行实际测试；游戏行业内的机器学习、计算几何和人机交互主题；同时研究 360-GPR 数据反演背后的物理和数学。