Real-time Rainfall Estimation Using Binarized Rain Streak Images in Surveillance Cameras

Abstract

<p>Real-time monitoring and analysis of rainfall are important in reducing potential damage and losses in water-related disasters. Nowadays, IoT sensor data is being widely used in weather observation because of cost-effectiveness with high spatiotemporal resolutions. This study proposes a novel approach to estimate rainfall intensity from binarized rain streak images in surveillance cameras. Here, several background subtract algorithms are considered to extract rain streak images from raw video data recorded by surveillance cameras installed in six different points in Seoul, Korea. Various ranges of binarization threshold values are also used to calculate the number of white pixel values from rain streak images. As results, it indicates that rainfall intensity is properly estimated from binarized rain streak images with a relation equation between the number of white values and observation rainfall intensity data, which shows high dependence on the amount of illumination and recording environment characteristics (e.g. rainfall type, camera parameter, etc.).</p><p>Keywords: Rainfall Estimation, Rain Streak, CCTV, Computer Vision, Korea</p><p><strong>Acknowledgement</strong></p><p>This work was funded by the Korea Meteorological Administration Research and Development Program under Grant KMI2022-01910 and in part supported by the National Research Foundation of Korea(NRF) grant funded by the Korea government(MSIT) (No. NRF-2022R1A4A3032838).</p>