Extraction of Urban Waterlogging Depth from Video Images Using Transfer Learning

Abstract

Urban flood control requires real-time and spatially detailed information regarding the waterlogging depth over large areas, but such information cannot be effectively obtained by the existing methods. Video supervision equipment, which is readily available in most cities, can record urban waterlogging processes in video form. These video data could be a valuable data source for waterlogging depth extraction. The present paper is aimed at demonstrating a new approach to extract urban waterlogging depths from video images based on transfer learning and lasso regression. First, a transfer learning model is used to extract feature vectors from a video image set of urban waterlogging. Second, a lasso regression model is trained with these feature vectors and employed to calculate the waterlogging depth. Two case studies in China were used to evaluate the proposed method, and the experimental results illustrate the effectiveness of the method. This method can be applied to video images from widespread cameras in cities, so that a powerful urban waterlogging monitoring network can be formed.