Non-Invasive River Level Measurement Using Trainable Segmentation in Computer Vision (#908)

Authors

Machado Olivares, Federico José

Aguilar Marroquín, José Alberto

Abstract

The importance of monitoring river levels in sectors of San Salvador has recently grown due to the increase in intense rains in recent years, particularly in the Tomayate River, located in the municipality of Apopa. Previously, the same authors have carried out research work that allows us to know the level of a high-risk river in the metropolitan area using computer vision, but the flood filling technique used, provides a very limited scope because it causes many errors when there is turbulence. In this work, the accuracy of level measurement has been significantly improved by applying a trainable segmentation technique, which provides a measurement error of less than 5% in about 68% of the processed images. This procedure will improve the development of early warnings of possible river flooding, thus reducing the risk of loss of material goods and human lives for those who live near the river.