A hybrid approach to estimate the complex motions of clouds in sky images. (15th November 2016)

• Record Type: Journal Article
• Authors: Peng, Zhenzhou; Yu, Dantong; Huang, Dong; Heiser, John; Kalb, Paul
• Is Part Of: Solar energy. Volume 138(2016)
• Page Start: 10
• Page End: 25
• Publication Date: 2016-11-15
• Subjects: Sky imagery -- Cloud motion tracking -- Optical flow; Solar energy -- Periodicals; Solar engines -- Periodicals; 621.47
• DOI: 10.1016/j.solener.2016.09.002
• Languages: English
• ISSNs: 0038-092X

Abstract

Highlights: Propose a hybrid model of cloud motion tracking based on block matching and optical flow. Extract dominant motion patterns to remove outliers and constrain motion field. Formulate a new optical flow energy-like objective to track deformation of clouds. Evaluate motion estimation models in both simulated and real image datasets. Abstract: Tracking the motion of clouds is essential to forecasting the weather and to predicting the short-term solar energy generation. Existing techniques mainly fall into two categories: variational optical flow, and block matching. In this paper, we summarize recent advances in estimating cloud motion using ground-based sky imagers and quantitatively evaluate state-of-the-art approaches. Then we propose a hybrid tracking framework to incorporate the strength of both block matching and optical flow models. To validate the accuracy of the proposed approach, we introduce a series of synthetic images to simulate the cloud movement and deformation, and thereafter comprehensively compare our hybrid approach with several representative tracking algorithms over both simulated and real images collected from various sites/imagers. The results show that our hybrid approach outperforms state-of-the-art models by reducing at least 30% motion estimation errors compared with the ground-truth motions in most of simulated image sequences. Moreover, our hybrid model demonstrates its superior efficiency in several real cloud image datasets by lowering at least 15% Mean Absolute Error (MAE) between predicted images and ground-truth images.