Flooding at low-water crossings, commonly referred to as “Irish crossings,” poses a significant risk to public safety, particularly in urban and semi-urban areas that are susceptible to flash flooding and sudden water surges. Traditional flood monitoring methods, reliant on manual inspection, lack the responsiveness required for real-time hazard mitigation. This paper proposes a cost-effective, real-time flood monitoring and management system that leverages Internet of Things (IoT) technologies, and computer vision to enhance flood detection accuracy. The system is built on a U-Net deep learning model for pixel-level water segmentation, achieving a segmentation accuracy of 92.36%. Integrated with a single-board microcomputer, the model processes image data at edge, enabling rapid detection of hazardous water levels and automated activation of safety mechanisms. This edge-based processing eliminates dependency on cloud connectivity, reducing latency and ensuring reliability even in remote or connectivity-limited areas. By addressing the limitations of traditional methods, the proposed system provides a scalable, real-time solution for proactive flood management, reinforcing infrastructure resilience and enhancing public safety.