雙光源系統於自動化鋪面檢測之研究

A Dual Lights Inspection Method for Automatic Pavement Survey

蘇詠順(Yung-Shun Su)

指導教授 ： 康仕仲

國立臺灣大學/工學院/土木工程學研究所/碩士(2010年)

https://doi.org/10.6342/NTU.2010.03313

摘要

鋪面檢測是項費時的工作， 但為了提供優質的用路環境，進行鋪面檢測是有其必要性的。許多的調查人員透過影像處理的方法提升檢測的品質，但是路面上的油漬、陰影及道路標示卻時常被誤判為道路破損，因此，在這個研究中，我們提出了雙光源檢測法(DLI)來減少誤判的機率。

雙光源檢測法主要有四個步驟：(1) 影像擷取- 我們利用不同的光源設定，從同一個位置及角度取得兩張影像。 (2) 影像相減- 我們將兩張影像中的每一個面對點相減用以找出兩影像中的不同點。 (3) 影像增強- 我們透過邊緣偵測取出影像中破損的特徵。 (4) 影像分類- 最後後我們使用分類演算法來認定影像中是否存在道路破損。

我們在道路上蒐集了212對影像用以對雙光源檢測法進行驗證，其中包含了42 對鱷魚狀裂縫、42對人首孔、58對縱向裂縫、34對表面髒污以及52對路面標示。我們從這些影像中取出20%(45對)建立分類模型，接著我們用剩餘的影像測試這個分類模型的準確度。我們將雙光源檢測法與傳統影像處理檢測法進行了比較後發現，雙光源檢測法可以明顯提升表面髒污(傳統：18%，DLI：82%)及道路標示(傳統：8%，DLI：96%)的辨別能力。用以辨別其它的道路破損一樣有不錯的準確度，包含有鱷魚狀裂縫(傳統：95%，DLI：90%)、人首孔(傳統：97%，DLI：100%)、縱向裂縫(傳統：62%，DLI：69%)。這個結果證實了雙光源檢測法是一個可靠度高的鋪面檢測法。

關鍵字

雙光源檢測 ； 成對影像 ； 鋪面檢測 ； 鋪面破損 ； 電腦視覺 ； 影像增強 ； 影像分類

並列摘要

A pavement survey is a time-consuming but necessary task to ensure the serviceability
of road pavements. Many investigators have used image-processing methods to automate
the survey processes and enhance the quality and accuracy of survey results. However,
the image-processing methods often mistakenly treat the oil spillages, shadow, and road
marking as distresses since their features are very close to the ones of distresses. Therefore,
in this research, we have developed a dual light inspection (DLI) method to reduce
false alarms.

DLI includes four major steps: (1) image capture: we retrieved two images as a pair
from the same position and orientation with two different light setups; (2) image subtraction:
we subtracted these two images pixel by pixel to obtain a subtracted image which
shows the differences between them; (3) image enhancement: we applied an edge detection
method to retrieve the distress features; (4) image classification: finally, we used a
classification algorithm to identify images including distresses with the ones without.

A field test was conducted to verify the DLI method. We took 212 pairs of images
at night, including alligator cracks (42 pairs), manholes (42 pairs), longitudinal cracks
(58 pairs), spillages (34 pairs), and road markings (52 pairs). Twenty percent of the images
(i.e. 45 pairs) are used as training sets to train the classification model. We then used remaining images to test the accuracy of the classification model. We compared
the accuracy between the DLI method, which uses dual light image pairs, and traditional
method, which uses individual images. We found that DLI can significantly improve the
accuracy in identifying spillages (traditional method: 18%, DLI: 82%) and road markings
(traditional method: 8%, DLI: 96%). The accuracy of the two methods in identifying
other distresses, including alligator cracks (traditional method: 95%, DLI: 90%), manholes
(traditional method: 97%, DLI: 100%) and longitudinal cracks (traditional method:
62%, DLI: 69%) was approximately. This result indicates that DLI is a reliable method
to conduct pavement inspections.

並列關鍵字

dual-light inspection ； pairing images ； pavement survey ； pavement distresses ； computer vision ； image enhancement ； image classification

參考文獻

[1] HD Cheng and M. Miyojim. Automatic pavement distress detection system. Information Sciences, 108(1):219–240, 1998.

[2] K.A. Abaza and I.A. Al-Khatib. Integrated pavement management system with a Markovian prediction model. Journal of Transportation Engineering, 130:24, 2004.

[3] Q. Li, M. Yao, X. Yao, and B. Xu. A real-time 3D scanning system for pavement distortion inspection. Measurement Science and Technology, 21:015702, 2010.

[4] Stephen G. Ritchie. Digital imaging concepts and applications in pavement man- agement. Journal of Transportation Engineering, 116(3):287–298, 1990. cited By (since 1996) 6.

[5] K.C.P. Wang and X. Li. Use of digital cameras for pavement surface distress survey. Transportation Research Record, 1675(1675):91–97, 1999. cited By (since 1996) 3.