透過您的圖書館登入
IP:3.137.178.133
  • 學位論文

影像辨識技術建立鋪面破壞三維模型之研究 -以坑洞破壞為例

Application of Image Recognition Study into Pavement Distress 3D-model Construction - Potholes

指導教授 : 林志棟 陳建達
若您是本文的作者,可授權文章由華藝線上圖書館中協助推廣。

摘要


巡查為道路養護的基本工作,鋪面破壞易造成用路人行車安全,其中又以坑洞破壞最為直接影響用路人之行車安全,國內因道路破壞造成人員傷亡之事件層出不窮,近年國內外均有使用影像攝影設備輔助道路巡查作業,降低其事故發生率,本研究以影像為基礎,以空間資訊量測之方法理論及影像處理技術,對道路破壞資料進行影像擷取蒐集並進行量測,再以ASTM D6433-11將鋪面破壞之類型進行評級判定。 本研究使用Bumblebee® XB3 CCD Camera雙基準立體視覺攝影系統進行開發,以空間資訊量測方法-空間前方交會法與影像處理技術-尺寸不變特徵轉換,以空間座標系、相片座標系、影像座標系之座標關係進行轉換,確立關係後,再將Bumblebee® XB3 CCD Camerar擷取之影像,以自製校正板進行相機律定及影像校正,可對道路破壞狀況進行量測,亦可將其資料進行道路破壞三維模型建立,藉以瞭解道路狀況。 本研究主要以道路坑洞破壞為實驗樣本,除可顯示破壞之三維資訊,於三維模型建立上較易進行判別,故以坑洞樣本為最合適,實驗結果則以人工量測與軟體量測進行比較,坑洞直徑量測誤差在5%以內,誤差最大值為1.1cm;深度量測誤差在7%內,誤差最大值為0.3cm,此外,人工與軟體PCI評級結果之顯著性為0.539大於0.05,顯示其評級結果並無顯著差異。未來可將此技術導入現今道路巡查作業流程,可減少於道路巡查作業耗費之時間及人力,再將蒐集之道路資料匯入鋪面維護管理系統,可做為未來鋪面工程生命週期評估及道路機關養護決策參考因子之一。

並列摘要


Road inspection is an essential work of road maintenance, pavement distresses would easily cause the lack of road safety. Among all the issues potholes affects the most. In recent years, the road maintenance authorities start to use the photographing devices and apply into the road inspection in order to reduce the time of road inspection and the accident rate. The study is based on the Theory of Spatial Information Measurement and computer vision image processing technology, through the pavement image capturing and pavement distress measuring to achieve the research purposes. It use Bumblebee® XB3 CCD Camera designed by the stereo vision system and use the spatial information measuring method –Space Intersection and image processing method - Scale Invariant Feature Transform. Through the spatial, photo, image coordinates relation conversion to calculate the real pavement distresses information and construct the 3D-model of pavement distresses. The study mainly take the potholes as the experimental sample, it can show the 3D information (length, width, depth) and easily determine the extent of distress with 3D-model.Comparison of the manual and program results, it shows that the biggest percentage error of the diameter and depth is 4.6% and 6.78%, the biggest error value is 1.1cm and 0.3cm. Besides, the comparison of manual and program pavement condition rating results, it shows that the distinctiveness is 0.539 lager than 0.05. Thera are no differences between the two rating method. In the future, the road maintenance authorities can apply the image recognition technology into the inspection and the road data can help the policy decision of the road maintenance.

參考文獻


李釗、陳繼藩、黃書猛、許峻榕、許書王(2002), Development of an Automated Airport Pavement Image Collection System,機場鋪面自動化影像蒐集系統之研究. Journal of Marine Science and Technology,10(1),1-7。
李柏(2012),鋪面剖面掃描儀應用於鋪面狀況辨識與偵測之研究,碩士論文,國立台灣大學土木工程學系。
林志勇、吳健碩(2005),鋪面破壞影像辨識系統之研究,鋪面工程,3(1),65-78。
黃祐祥(2010),多重影像匹配於房屋模型重建,碩士論文,國立中央大學土木工程學系。
ASTM D6433-11(2011) Standard Practice of Roads and Parking Lots Pavement Condition Index Surveys.

延伸閱讀