Date of Award


Document Type

Open Access Thesis

Degree Name

Master of Science (MS)


Computer Science

First Advisor

Marc Pomplun

Second Advisor

Craig Yu

Third Advisor

Dan Simovici


Many communities along coastlines and riverbanks are threatened by water erosion and hence an accurate model to predict erosion events is needed in order to plan mitigation strategies. Such models need to rely on readily available meteorological data that may or may not be correlated with the occurrence of erosion events. In order to accurately study these potential correlations, researchers need a quantified time series index indicating the occurrence and magnitude of erosion in the studied area. We show that such an index can be obtained by creating and analyzing a single image series using relatively cheap consumer grade digital cameras. These image series are naturally of lower quality and subject to a large amount of variability as environmental conditions change over time. We initially analyze each image as a whole and subsequently demonstrate the great advantages of segmenting each image. This allows for independent parallel processing of segments while preventing cross-contamination between them. Finally, we are able to automatically detect 67% of all erosion events while accepting only a small number of false positives.