Page 193 - 中華技術127期

摘要 摘要
橋梁檢測最基本檢測方式為目視檢測，由檢測人員以目力巡視橋梁構件外觀，就其損壞狀況逐一拍 照與記錄，並依其所受之專業訓練、經驗進行評等分級。然而上述程序所拍攝之照片沒有量測與定位能 力，且難以達到事後檢核與變異偵測的目的。此外，若遇到人員不易到達或無法以近距離觀測的橋梁， 例如跨河橋或跨山谷橋，則需採用人員吊掛或橋檢車近距離檢測，容易造成人員安全與經費工時增加的 問題。目前世界各國公路維護單位已逐步採用無人機或機器人協助拍攝影像，以達到間接目視檢測的目 的。由於大量的影像，透過人眼逐一檢視，將是繁重費時的工作。因此，本研究提出影像式智慧橋梁檢 測作業程序，讓操作人員透過視窗操作軟體在室內進行橋梁檢測，包括使用人工智慧機器學習技術從大 量影像中自動判識劣化區，並透過攝影測量技術協助三維定位，以達到多時期劣化區變異偵測與量化的 目的。本研究所建議之標準作業程序經過五座2-3跨混凝土橋梁的實際測試，驗證本研究所提出之作業程 序，對視覺式橋梁檢測已可達到相當高的實用性。
12345
Development of an image-based smart bridge
Abstract
inspection system
Abstract
The fundamental method of bridge inspection is the human visual inspection, i.e. the inspector has to exam the bridge surface visually. If any deterioration has observed, then take photo for recording and judging its damage level according to the professional training and experience. However, the photos taken by the above procedure are not suitable for measurement and positioning, particularly not capable of posterior quality check and change analysis. In the meantime, for bridges that were constructed above river or valley are difficult to reach its bottom for close-range inspection, then the inspector may need to be hanging by a rope or carrying by a bridge inspection vehicle. It will introduce higher risk to human life and increase the cost and work time. Therefore, most of the road maintenance agency in the world are trying to utilize unmanned aerial vehicle (UAV) or robot to take images and indirectly assist the bridge inspection. However, it is tedious and time consuming to inspect the acquired massive images one-by-one by human eyes. Thus, in this study we suggest an image-based smart bridge inspection procedure that allow the inspector to check the bridge surface within the office through a windows-based software. It includes the use of artificial intelligent method through machine learning to recognize the deteriorated area from the acquired images automatically, determining the position of the deteriorated area through photogrammetric techniques, and achieve the purposes of multi-temporal change analysis and quantization. The proposed operational procedure has been verified by 5 concrete bridges with 2-3 sections long. It proves that the proposed procedure can reach high feasibility to vision-based bridge inspection.
4
No.127│July, 2020│ 191
特 稿