以無人機建立即時自動化橋梁裂縫影像辨識系統

Abstract

在建築物和橋梁的生命週期中，使用階段佔了絕大部分的時間和資源，這包括了後續的維護和定期檢修，因此要如何更有效地管理和監測它們成為一個極為重要的議題。傳統的檢測方法主要依賴目視檢測，這需要花費大量的時間和人力成本。尤其在橋梁檢測方面，橋檢人員為了監測特定區域，必須承受高安全風險。而在建築物檢測方面，通常需要監測有倒塌風險的建築物，這使得監測人員面臨極高的生命風險，因此，如何改善這些監測方法成為一個迫切需要解決的問題。 近年來，隨著 AI (Artificial Intelligence) 相關技術的發展，尤其是深度學習演算法的興起，目前已經許多研究學者將無人機 (Unmanned Aerial Vehicle, UAV) 和影像辨識結合，利用無人機的高機動性以解決因位置不便而橋檢人員難以到達的問題，或是無人機的大面積偵查能力提高工作效率。然而，過去的研究大多僅將無人機用於現場拍攝，再將所拍攝的影像做後續的影像處理和辨識，這樣需花費額外人力和時間成本，而且導致缺乏即時性，使用者無法即時獲取影像資訊。 隨著 ROS (Robot Operating System) 系統的蓬勃發展，這一問題迎刃而解， ROS 系統為軟硬體整合提供了一個便利的平台，藉此，無人機影像辨識就能夠實現即時化，從而進一步提升了監測效率和即時性。本研究旨在結合無人機和影像辨識模型，透過 ROS 系統建立一個即時的裂縫辨識系統。 該系統首先將無人機捕獲的影像進行一系列的影像處理步驟，最終將處理完的影像，匯入已訓練好的影像辨識模型進行裂縫辨識。辨識結果將被儲存，同時向使用者發出警示通知，實現即時的裂縫檢測和通報功能。這樣的整合機制不僅提高了工作效率，也確保了即時性和準確性，有望未來能在建築物及橋梁的監測和維護中發揮重要作用。 除此之外，該系統還可應用於災後搜救和災害評估，提高搜救工作的效率和準確性。本研究旨在建構一個即時的裂縫影像辨識系統，以提高建築物和橋梁監測作業的效率和安全性，以應對日益嚴峻的社會需求。

In the life cycle of buildings and bridges, the operational phase occupies the majority of time and resources, including subsequent maintenance and periodic inspections. Effective management and monitoring of these structures have become crucial issues, given the substantial investment required. Traditional inspection methods primarily rely on visual assessments, which demand significant time and human resources. Particularly in bridge inspections, personnel face high safety risks when monitoring specific areas. Similarly, building inspections often require monitoring structures at risk of collapse, posing severe life-threatening risks to the inspectors. Thus, improving these monitoring methods is an urgent problem that needs to be addressed. In recent years, with the development of Artificial Intelligence (AI) technologies, especially the rise of deep learning algorithms, many researchers have begun integrat- ing Unmanned Aerial Vehicles (UAVs) with image recognition to leverage UAVs’ high mobility for addressing the difficulties inspectors face in accessing certain areas, or to enhance work efficiency through UAVs’ extensive reconnaissance capabilities. However, past studies mostly utilized UAVs for onsite shooting, followed by subsequent image pro- cessing and recognition, which incurs additional labor and time costs and lacks real-time capabilities, preventing users from obtaining immediate image information. With the rapid advancement of the Robot Operating System (ROS), this issue can now be effectively resolved. The ROS provides a convenient platform for hardware and software integration, enabling real-time image recognition through UAVs, thereby further enhancing monitoring efficiency and immediacy. This study aims to develop a real-time crack recognition system by integrating UAVs and image recognition models through the ROS. The system first processes images captured by the UAV through a series of image processing steps. The processed images are then input into a pre-trained image recognition model for crack detection. The recognition results are stored and simultaneously send alert notifications to users, achieving real-time crack detection and reporting. This integrated mechanism not only improves work efficiency but also ensures immediacy and accuracy, promising significant contributions to the monitoring and maintenance of buildings and bridges in the future. Additionally, the system can be applied in post-disaster search and rescue and disaster assessment, enhancing the efficiency and accuracy of rescue operations. This study aims to construct a real-time crack image recognition system to improve the efficiency and safety of building and bridge monitoring operations, addressing the increasingly severe societal demands.