발간년도 : [2023]
| 논문정보 |
|
| 논문명(한글) |
[Vol.18, No.5] Pothole Detection Using Deep Learning and Domain-based Image Preprocessing Methods |
|
| 논문투고자 |
Jin-Nyeong Heo, Yeong-In Lee, Ha-Young Kim |
|
| 논문내용 |
In recent years, the number of potholes has been increasing due to the high frequency of heavy rainfall. In addition, road surface damage is inevitable due to the aging of roads, and damaged roads (potholes and cracks) interfere with drivers' driving, causing various safety accidents. To efficiently solve this problem, various road damage detection studies have been conducted using artificial neural networks. However, previous studies have been limited by a lack of understanding of potholes. Furthermore, the need for research is growing as the number of risk factors that can cause potholes is rapidly increasing. To overcome the limitations of previous studies, this study proposes an image preprocessing technique that can effectively reflect the characteristics of potholes and an optimal structure for pothole detection based on EfficientDet. The proposed preprocessing technique combines two algorithms, CLAHE and Sobel Edge detection, to identify and learn potholes in road surface images by maximizing the boundaries of potholes through contour detection and contrast thresholding for the entire image, rather than solely relying on contrast. In addition, we designed the optimal number of BiFPN layers for the pothole dataset so that the module can clearly detect potholes. The methodology proposed in this study was applied to EfficientDet and YOLO v5 models to experimentally prove the feasibility of the methodology. |
|
| 첨부논문 |
|
|
|
|
|
자료실 > 논문지 
