3D Cloud Position Estimation for Solar Radiation Forecasting Using Time-Series Stereo Vision Integrated with All-Sky and Satellite Images

Minoru Harada, Yusuke Kameda

Abstract: We propose a method to estimate the 3D position of clouds using time-series stereo vision from all-sky and satellite images, aimed at improving the accuracy of short-term solar radiation forecasting for photovoltaic power generation. Our method focuses on the temporal dynamics of cloud movement, eliminating the need for pattern matching across images from different viewpoints. By capturing a sequence of all-sky images and using optical flow, cloud motion on the celestial sphere can be tracked. Additionally, by utilizing the cloud motion speed on satellite images and assuming that the vertical movement of the cloud is much smaller compared to their horizontal movement, we estimate the 3D position. We evaluate the effectiveness of our approach using both simulated and real cloud videos. Our results demonstrate that the proposed method can estimate the 3D positions of various types of clouds, including thick and thin clouds, expanding the range of cloud types that can be effectively analyzed. Quantitative evaluations confirm the method's robustness.