Detecção automática do nível de água por imagens utilizando visão computacional

Abstract

Monitoring and management of water level has become an essential task in hydroelectric powers. Activities as water resources planning, supply basin management and flood forecasting are mediated and defined through its monitoring. The measuring station is located on the Madeira River in Rondônia, Brazil. The measurements are made by sensors installed on the river facilities that carry out the estimations precisely. Weather conditions influence the results obtained by these sensors; therefore, it is necessary to have redundant approaches besides the sensors which maintain the high accuracy of the measured values. Conventional cameras supply this necessity measuring it through human eyes. However, its method is not reliable and has low accuracy. It is proposed an approach, in redundancy to sensors, to use image processing to measure the water level, with high accuracy, and to predict the water level measured using a convolutional neural network for regression. Results present low errors according to its prediction.