AI-CCTV Model for Real-time Flood Monitoring: Outland River Discharge and Inland Flood Depth Estimation

Recently, flood damages have increased due to torrential rains driven by climate changes. In order to overcome the limitation of traditional manpower-dependent river monitoring approaches (low level of automation and system rapidity), CCTV-based digitalized monitoring techniques should be developed for promptly monitoring, predicting, and responding to flood damages. This study introduces an integrated AI-CCTV model for both outland and inland flood monitoring in real-time: Optical-Flow-based Flow Estimation Model (OF2EM) for outland river discharge estimation and Multi-Object-based Segmentation for Flood Depth Estimation Model (MOSDEM) for inland flood inundation. The two sub-modules of the integrated model were demonstrated independently to each domain, (1) the former OF2EM in a laboratory flume with various hydraulic and optical conditions (e.g., different brightness contrast, optical interference, with hydraulic structure that generates eddy flows) and (2) the latter MOSDEM to identify each different group of objects in a CCTV frame and finally to quantify the extent and depth of inundated area. Finally, the proposed integrated AI-CCTV model is verified to confirm the benefit of integrating the dual sub-models of real-time flood monitoring (e.g., inundated water propagation prediction with OF2EM in one corner of a frame and inundation boundary and severity determination with MOSDEM in another corner of the frame).