[A-HW24] Human- and Climate-induced variability in water cycle and (sub)surface water resources

[AHW24-P01] Development of Real-Time Water Level Detection Technique by CCTV and Deep Learning

*Zheng-Da Jiang¹, Yan Zhuang¹, Yuan-Chien Lin¹ (1. Department of Civil Engineering,National Central University)

Keywords:Deep learning, Image detection, Water level detection, CCTV

With the intensification of climate change, the frequency of extreme rainfall has been increasing, leading to a heightened risk of urban flooding and river overflows. As a result, real-time monitoring of water level changes has become a crucial measure for disaster prevention and response. Traditional water level monitoring methods primarily rely on water gauges and sensors; however, their widespread adoption is often hindered by installation costs, maintenance requirements, and environmental constraints. To enhance the accessibility and feasibility of real-time water level monitoring, this study explores the application of Closed-Circuit Television (CCTV) in real-time water level monitoring, utilizing image processing and deep learning techniques to analyze water level fluctuations. The system aims to provide real-time water level information as a reference for floodgate control, ensuring effective water resource management and utilization. In this study, a Mask R-CNN model is developed to identify water areas in CCTV-captured images. The system calculates water level variations by analyzing the vertical changes in water pixels at designated points and applying time-series analysis. When water level fluctuations exceed a critical threshold, the system can issue early warnings, assisting relevant authorities in responding promptly to reduce flooding risks and minimize potential disasters.