Image Recognition of River Water Gauges Using Polynomial Regression Model for Predicting Binarization Threshold

Abstract

Taiwan is frequently affected by typhoons. Typhoons bring heavy rainfall and cause rapid river level rise and even flooding. In the past, high-accuracy but costly equipment, which could not be widely distributed, was used for hydrological observation. It caused us unable to obtain regional hydrological information in real-time. Currently, CCTV has been widely distributed in rivers in Taiwan, and real-time weather information can be accessed through the Internet by the public. Using CCTV and public weather information, we analyzed the images of the water level gauge to provide real-time regional hydrological information for early flood warnings. The image binarization is used for the analysis of water levels. However, because of the differences in environmental factors such as time, weather, and sunrise/sunset time, a set of different thresholds must be used for the binarization in image processing. In this study, a polynomial regression model for predicting the binarization threshold was proposed. According to the changes in environmental factors, the threshold required for image binarization was predicted in real-time, thereby improving the image recognition rate of water gauges.