應用數位影像檢測道路標線可見度

Application of digital image on road marking

梁健偉(Kin-Wai Leong)

指導教授 ： 周家蓓

國立臺灣大學/工學院/土木工程學研究所/碩士(2019年)

https://doi.org/10.6342/NTU201902237

摘要

因應目前國內外對道路安全的重視程度以及要求都在不斷的提高，而道路標線作為道路的重要組成部份，相關之驗收及檢測標準目前卻非常缺少。目前國際上常用於標線反光度的評估指標主要有兩個：一為擴散照明下之輝度係數(Qd)，另一為回歸反射輝度係數(RL)，然而該反光度指標的量測方法以及要求較高，難以在一般道路上進行廣泛且具有恆常性之檢測，也因此導致道路上的不同標線反光度能力參差不齊。有鑑於此，本研究提出運用車載相機進行影像辨識建立具有鑑別效果的標線反光能力檢測系統，應用Mask R-CNN模型找出影像中標線所在的位置，把該位置中的像素點亮度取平均作為該標線之照片亮度，並使用標線在影像之位置利用雙鏡頭之視差應用前方交會法進行距離測量，最後以上述之數據與標線反光度指標RL進行比較，訂定一個能夠分辨標線合乎最低標線反光能力要求之照片亮度值作為門檻值，以供日後標線養護單位養護之參考。研究建立的標線照片亮度值與標線反光度指標RL在不同分級的趨勢具有一致性，能夠相當的反映標線的反光能力以及可視程度。
而本研究之另一部份為受試者實驗，用以探討駕駛者在實際駕駛時對標線之可視狀況與影像辨識之數據差異、不同標線類型、標線所在之鋪面類型、車輛行進狀況、外部光源對受試者之直接影響等等。而根據實驗結果顯示受試者駕駛時標線之可視狀況基本都比影像辨識要來得好，不同標線之可視距離與RL值大小對應相乎，但不同年齡層、車輛行進狀況對於標線之可視距離的影響都不大，本研究為首次利用雙鏡頭影像取得標線夜間之照片亮度值與反光能力關係，為道路管理單位在使用中標線之反光性能檢測功能奠定新的發展方向。

關鍵字

道路標線 ； 反光度 ； 影像辨識 ； 雙頭鏡 ； 受試者實驗

並列摘要

In view of the current domestic and international emphasis on road safety and the requirements are constantly improving, road markings as an important part of the road the relevant acceptance and testing standards are currently very lacking. At present, there are two main evaluation indexes commonly used in the international standard for the grading of road marking: one is the luminance coefficient under diffuse illumination (Qd) and the other is coefficient of retroreflected luminance(RL). However, the measurement method of those index and requirements are high, and it is difficult to carry out extensive and constant detection on the general road, and as a result, the reflective capability of different markings on the road is uneven. In view of this, our research proposes to use the on-board camera for image recognition to establish a road marking reflective capability detection system with discriminative effect. Mask R-CNN model is used to find the position of the target road marking in the image, and the brightness of the pixel in the position is taken, after that the average brightness of those pixel is used as the brightness of the road marking. The maximum visible road marking distance is measured by using the different position of the pixel for a same marking in the dual lens to apply the front intersection method. Finally, compare the above data with the coefficient RL, set a photo brightness value that can distinguish the road marking from the minimum road marking's reflective capability as a threshold for future road marking maintenance reference. The brightness value of the road marking photograph established by the study is consistent with the trend of the grading index RL in different grading, which can reflect the reflective capability and visibility of the road marking.
The other part of the research is a subject experiment to explore the data difference between the visual condition and the image recognition of the driver during actual driving, such as: different types of markings, type of pavement where the marking is located, the state of vehicles, impact of external light sources on the subject, etc. According to the experimental results, the visual condition of the subject's driving the road marking is basically better than the image recognition. The visual distance of different markings corresponds to the RL value, and the visual distance of the flexible pavement is generally It will be higher than the rigid pavement. However, in different age groups, the vehicle's traveling condition has little effect on the visual distance of the marking. But if there is an external light source, the visible distance of the material selected in this study exceeds the minimum reaction time required under the general road speed limit, and the brightness of the photo of the road marking is also meets with the system's analysis exceeds the threshold set by the system.

並列關鍵字

Road Marking ； Reflective Capability ； Image Recognition ； Dual lens ； subject experiment

參考文獻

[1] AASHTO M247-13. Glass Beads Used in Pavement Markings. American Association of State Highway and Transportation Officials, Washington DC, USA. (2013).

[2] Abdulqadir, Omar Y. "Distance Measurement Using Dual Laser Source and Image Processing Techniques."Al-Rafidain University College For Sciences35 (2015): 266-286.

[3] Ahmad, Touqeer, et al. "Symbolic road marking recognition using convolutional neural networks."2017 IEEE Intelligent Vehicles Symposium (IV). IEEE, 2017.

[4] Burns, David M., Thomas P. Hedblom, and Terry W. Miller. "Modern Pavement Marking Systems: Relationship Between Optics and Nighttime Visibility."Transportation Research Record 2056.1 (2008): 43-51.

[5] Babic, Darko, Mario Fiolic, and Petar Prusa. "Evaluation of Road Markings Retoreflection Measuring Methods." European Scientific Journal, ESJ 10.7 (2014).