|
Title |
A Study on Estimation of Flood Area in Urban Areas Based on Computer Vision
|
|
Authors |
김승우(Seungwoo Kim) ; 최웅규(Woonggyu Choi) ; 나상일(Sangil Na) ; 박승희(Seunghee Park) |
|
DOI |
https://doi.org/10.11112/jksmi.2025.29.6.49 |
|
Keywords |
도시 침수; 컴퓨터비전; 면적 추정; 실시간 감지; 딥러닝 Urban Flood; Computer Vision; Area Estimation; Real-time Detection; Deep Learning |
|
Abstract |
The acceleration of climate change has increased the frequency of localized heavy rainfall and torrential rain, raising the risk of flooding in urban areas. Existing flood monitoring methods based on manual surveys and sensors have limitations such as a restricted application range, high operation, maintenance, and management costs, and a lack of real-time capability. This study introduces a computer vision?based method capable of detecting urban flooded areas in real time and quantitatively estimating flooded areas. A YOLOv8-based deep learning model is used to simultaneously detect flooded areas, traffic signs, traffic lights, and the actual size of the reference objects combined with pixel information in the image is utilized to calculate the flooded area in square meters. The training dataset was constructed by combining data collected from AI-Hub, and the final model achieved a mAP of 0.548 and an f1-score of 0.59. In addition, a field experiment conducted in areas frequently affected by flood showed an error rate of approximately 5.63% compared to the area estimation based on satellite imagery, demonstrating high reliability.
|