Inundation monitoring using a machine learning algorithm combining AI and edge detection
Abstract
Advanced integration of passive remote sensors with cutting-edge machine learning algorithms and edge detection has expedited the potential for autonomous flood observation systems leveraging passive remote sensing. This research study field tested web ready video cameras capable of extracting free surface water levels from live oblique pictometry in tidal water bodies. Live-streaming web cameras were strategically deployed to collect images in 6 min intervals over a 3-month period. Statistical analysis demonstrated that these sensors effectively provided continuous surface water level measurements, with a RMSE deviation of <1.25 cm when positioned within approximately 10 m of the desired monitoring area. To visually verify vertical accuracy, A-style staff gages were installed in the camera's field of view, and for each observation station, reported water levels in the captured imagery were cross-validated using nearby Ka-band radar sensors. Various camera hardware models were evaluated, each equipped with a high resolution optical sensor for optimal pixel density, and an integrated infrared projector to enable monitoring in the dark within a confirmed range of <30 m. Ultimately, this study developed and verified the potential of a novel passive remote sensing web camera, using still frame images from online web cameras to accurately detect and extract free surface water elevations from time-lapsed videos.