Depth Estimation in Urban Flooding Using Surveillance Cameras and High-Resolution LiDAR Data

Abstract

Urban flooding disrupts transportation networks, making accurate real-time flood depth estimation on roads crucial. At regional scales, flood extents from satellite imagery are overlaid with Digital Elevation Models (DEMs) to estimate flood depths. However, at street scales, flood extent images are captured by surveillance cameras, which provide perspective rather than top-down views. In this paper, we present a reliable method for estimating flood depth at street scales by integrating high-resolution DEMs, obtained via LiDAR, surveillance camera imagery, and pinhole camera models to project the DEM onto the image. Our algorithm generates a series of “artificial floods” by truncating the projected DEM at different elevations. The flood depth is then determined by maximizing the alignment between the artificial and observed flood extents. Validation using ground-truth data shows that our approach achieves an average error of less than 1 cm for flood depths above 5 cm, although performance declines for shallower depths.