A novel approach to urban flood monitoring using computer vision

Due to the high density of buildings and improper water drainage facilities, flash floods are prevalent in cities. In such scenarios, to extend timely aid, it is often required to estimate the level of flood in different parts of the affected zone. This paper presents a novel method employing participatory sensing and computer vision to estimate the flood level. The method involves the participants capturing and uploading images of the partially submerged static structures such as buildings, lampposts etc., using their smart phones or other intelligent devices. The captured images are geo-tagged and uploaded to a server. The feature matching algorithm, SIFT finds the corresponding matching feature points between the captured and a reference image at the server. The flood line is then estimated and drawn against the reference image. To optimize the network bandwidth usage, images with average resolutions are used. This method yields immediate and considerably accurate results thus helping in isolating areas that have been severely impacted for timely help.