Date of Award
Fall 2020
Document Type
Thesis
Degree Name
Master of Science (MS)
Department
Electrical and Computer Engineering
First Advisor
Medeiros, Henry
Second Advisor
Povinelli, Richard
Third Advisor
McDonald, Walter
Abstract
Watershed models of extreme flooding events require accurate and reliablemeasurements of streamflow for calibration and validation. However, flow rate measurements during floods are inherently uncertain, and physical measurements of velocity during flood conditions are prohibitive in many cases. Therefore, novel methods to measure stream velocity during extreme floods must be considered. This work addresses this challenge through the development of the Aerial Imaging Riverflow System (AIRS), a novel system that utilizes drones, aerial imagery, and optical flow algorithm to measure velocity in rivers and streams. This system was applied at a case study location on the Menomonee River in Wauwatosa, Wisconsin. To remotely sense streamflow, a DJI Matrice 210 RTK drone equipped with a Zenmuse X5S camera was used to capture video. River velocity was measured using a combination of point measurements captured with a hand-held velocity meter. The video data from the drone was analyzed using optical flow generated by the PWC-Net model (a neural network for optical flow estimation) to generate velocity estimates. Results indicate that on average the optical flow algorithms estimate the magnitude of surface velocity to within 13%-27% of hand-held measurements without the use of artificial seeding. This outcome suggests that this system could be used as an alternative to measure velocities in rivers and streams.