Image Segmentation with Zero-Shot Generalization for Flood Prediction in Urban Environments

Abstract

Accurate and continuous monitoring of river levels is crucial for managing water resources and preventing natural disasters. This project aims to analyze the water levels of urban creeks using images and develop a prediction algorithm based on the extracted information. The study seeks to explore the feasibility and effectiveness of using image analysis to evaluate, monitor, and predict the water level of the Mineirinho Creek in São Carlos, Brazil. The methodology involves regularly capturing images over a specific period and processing them using various computer vision techniques to determine the water level. The Segment Anything (SAM) model was employed to segment the water body, while a sequence of steps was adopted to provide a flooding index. A comparative analysis confirmed the accuracy and reliability of the proposed method.