Combining Unmanned Aerial Vehicles (UAVs) and close-range photogrammetry has become a safer, more efficient, and cost-effective solution for bridge inspection compared to conventional methods. However, close-range bridge images captured by UAVs often suffer from severe quality issues, such as blurriness, improper exposure, limited coverage, or insufficient resolution. These issues can adversely affect subsequent damage identification and bridge condition assessment, thereby hindering the widespread application of UAVs in bridge inspection. This paper presents a novel in-flight image quality check (IIQC) framework for rapidly assessing UAV-captured images against bridge inspection requirements. This framework incorporates (1) a new coarse-to-fine image pose estimation approach for precisely estimating the relative poses of UAV images with respect to a reference model, (2) a comprehensive set of refined image quality metrics for assessing image quality across various aspects, and (3) an Image Quality Metrics (IQM)-embedded bridge representation model for visualising the evaluation results. All components within this framework have been validated through extensive simulation and real-world experiments, encompassing various bridge structures (i.e., girder, arch, and truss bridges) and different weather conditions. The results show that: (1) the average root mean square error (RMSE) of the estimated image poses by the coarse-to-fine method across multiple structures reached 0.189 m in position and 0.203° in orientation, showcasing an improvement over 50 % than the image poses retrieved from the UAV; (2) the image quality metrics offer comprehensive insights into image blurriness probability and exposure intensity at the pixel level and can effectively identifying missing and insufficient captured areas; (3) the bridge representation model is capable of delivering valuable and user-friendly feedback of the assessment results to the pilot; (4) the IIQC framework can offer a rapid and comprehensive assessment of close-range UAV images in near real-time by leveraging an existing Building Information Model (BIM) of the target bridge, serving as a novel and efficient solution to thoroughly tackle image quality issues in UAV-enabled bridge inspection. Its versatility can potentially be extended to encompass other structural types, including buildings and dams.

中文翻译：

---

用于无人机桥梁检查的快速飞行图像质量检查


与传统方法相比，无人机 (UAV) 和近距离摄影测量相结合已成为更安全、更高效且更具成本效益的桥梁检测解决方案。然而，无人机拍摄的近距离桥梁图像经常存在严重的质量问题，例如模糊、曝光不当、覆盖范围有限或分辨率不足。这些问题会对后续的损伤识别和桥梁状况评估产生不利影响，从而阻碍无人机在桥梁检测中的广泛应用。本文提出了一种新颖的飞行图像质量检查（IIQC）框架，用于根据桥梁检查要求快速评估无人机捕获的图像。该框架采用了（1）一种新的从粗到细的图像姿态估计方法，用于精确估计无人机图像相对于参考模型的相对姿态，（2）一套全面的细化图像质量指标，用于评估各种不同环境下的图像质量。 (3) 嵌入图像质量指标 (IQM) 的桥表示模型，用于可视化评估结果。该框架内的所有组件均已通过广泛的模拟和真实实验进行了验证，涵盖各种桥梁结构（即梁桥、拱桥和桁架桥）和不同的天气条件。结果表明：（1）通过由粗到细的方法估计的图像姿态在多个结构上的平均均方根误差（RMSE）在位置上达到0.189 m，在0上达到0。方向为 203°，比从无人机检索到的图像姿态提高了 50% 以上； （2）图像质量指标提供了像素级图像模糊概率和曝光强度的全面洞察，可以有效识别丢失和不足的捕获区域； (3) 桥梁表示模型能够向飞行员提供有价值且用户友好的评估结果反馈； （4）IIQC框架可以利用目标桥梁现有的建筑信息模型（BIM），近乎实时地对近距离无人机图像进行快速、全面的评估，作为彻底解决图像问题的新颖而高效的解决方案无人机桥梁检测中的质量问题。其多功能性有可能扩展到其他结构类型，包括建筑物和水坝。