Absolute Urban Floodwater-Level Estimation from CCTV Imagery via Laser-Scanning-Based Image Projection

Abstract

Absolute estimation of urban floodwater levels from closed-circuit television (CCTV) imagery remains a challenge, particularly under nighttime illumination, reflections, and rain-induced noise, where segmentation-based and reference-object methods often degrade in performance. Although laser-scanning point clouds provide reliable 3D geometry, their integration with fixed urban CCTV streams for real-time, metric-scale flood monitoring remains insufficiently explored. This study presents a unified CCTV–laser-scanning framework for detecting flood events and estimating absolute floodwater levels in urban environments by integrating a time window–based incident detection scheme with an edge-based adaptive moving-window waterline tracking method. A two-stage photogrammetric calibration procedure—direct linear transformation (DLT) followed by single photo resection (SPR)—recovers camera orientation parameters for uncalibrated CCTV cameras, which enables the generation of image-projected point cloud data from terrestrial laser scanning (TLS) and mobile mapping system (MMS) measurements. Pixel-level water–structure contact lines extracted from CCTV imagery are subsequently projected onto the image-projected point cloud data, enabling metric-scale floodwater-level estimation. The proposed framework was validated under severe nighttime rainfall conditions using four urban CCTV data sets recorded during Typhoon Mitag (2019) in Uljin, South Korea. The results demonstrate reliable flood event detection and accurate floodwater-level estimation, with the latter achieving subdecimeter-level accuracy for both TLS- and MMS-derived point cloud data. These findings indicate that the photogrammetric integration of CCTV imagery with laser scanning offers a practical basis for real-time urban flood monitoring and disaster-response applications.