Image Processing for Flood Detection and Prediction

Several regions around the world, particularly cities in Brazil, have been suffering the consequences of flash floods in urban environments. Damage control measures include the deployment of sensors to identify the imminence of floods in critical areas. In this work, we investigate computer vision-based techniques to identify the imminence of floods based on images from cameras placed next to a river in São Carlos. We considered the raw images and dense optical flow computations from successive image collections and performed supervised learning with deep neural networks. Results show that the proposed approach is able to identify events of interest.