Unlocking Data from the Online Footage of the Edenville Dam Breach
Yixing Yuan, David Bazzett, Rishika Padnani, Ruo‐Qian Wang · Geo-Extreme 2021 · 2021
This paper presents a computer vision analysis to extract data from the footage of the Edenville Dam breach. The Edenville Dam breached on May 19, 2020, leading to extensive flooding in Central Michigan. Engineers have been eagerly seeking first-hand information to study this catastrophic event. Fortunately, the dam breach was captured by a local resident with a smartphone and was published online. Leveraging computer vision technology, we conducted an in-depth analysis of the video in the following steps: (1) geo-reference the 2D scene of the video with 3D real-world coordinate; (2) stabilize the camera movement and recover the dam motion; (3) track soil body motion via optical flow and interpret geotechnical failure mechanism; and (4) extract data for displacement and velocity and reconstruct 3D scene. The acquired data offer better visualization and insights of the dam failure mechanism. The proposed method has great potential in forensic analyses of geohazards’ events.