Comparative study of algorithms for cloud motion estimation using sky-imaging data

Abstract

Abstract One way to improve the efficiency of concentrated solar power plants is to use short-term forecasts of Direct Normal Irradiance (DNI) to apply advanced control strategies. These short-term forecasts are obtained using sky imagers along with signal and image processing techniques. An important step in DNI forecasting is cloud motion estimation. Block matching and optical flow algorithms are the most widely used techniques for motion estimation. In this paper, we present a comparative study between the various existing approaches used to estimate the velocity field of clouds. The comparative study is made using a database of images captured by a sky-imaging system developed at PROMES-CNRS laboratory. Computational time and estimation accuracy are the two criteria used to evaluate the performance of each algorithm. The results demonstrate the high performance of optical flow algorithms in terms of accuracy.