Automatic Detection of Flood Severity Level from Flood Videos using Deep Learning Models

摘要

Flood is a very commonly occurring disaster affecting a lot of people across the globe. Hence, flood risk assessment becomes a serious concern, which can reduce the damage or affects caused by the floods. This assessment can help in forecasting the flood situation, generating the early warning, handling the disastrous situation, providing the immediate help and performing the rescue operation in a flooded region. In this paper, we explore the use of deep learning models for predicting the severity level of a flooding event captured in videos by the habitats in a flooded region. The proposed model takes the video of a flooding event as input and determines its severity level. We also create a dataset of flood videos due to the unavailability of such special kind of dataset. The proposed model is then evaluated on this dataset and compared with a baseline convolutional neural network (CNN) based model. Simulation results reveal that the proposed model outperforms the baseline model in terms of accuracy for classification of videos of flooding events into severity levels.