Improving flood detection with large-scale dashboard camera data

In this paper, the authors developed a flood detection approach that applies computer vision to widely available dashcam imagery to identify street flooding, highlighting previously undetected floods and biases in existing urban flood detection methods. This approach harnesses the ubiquity of dashcam-generated dense street imagery to enhance flooding resilience, providing a cost-effective means to improve detection of flooded areas without the need for extensive manual labeling or sensor deployment. 

By identifying overlooked flooded neighbourhoods, quantifying biases in existing detection methods, and suggesting strategic locations for new flood sensors, this work has the potential to provide urban planners, policymakers, and residents themselves with useful insights into flood management and preparedness, complementing existing detection methods and supporting more targeted interventions and better response planning. Looking ahead, this work presents several promising directions for future research. Expanding the analysis to additional locations could help validate the approach in diverse contexts and provide new insights into local flooding patterns. Another promising direction is developing real-time flood monitoring capabilities to support more timely emergency responses; this would likely require deploying more computationally efficient models or edge computing methods directly onto networked dashcams