Sky analysis from CCD images: Cloud cover

Abstract

One of the tasks in sky modelling is the determination of the location and amount of cloud coverage in the sky. Analysis using digital camera systems is possible but requires an automated method of separating sky images into clear sky and cloud regions. Two segmentation approaches have been used in this research, one based purely upon the colour characteristics of sky regions, the other a neural network approach using a wider range of variables. A convolution technique was developed to reduce classification errors prior to defining cloud outlines using polylines. Sensitivity analysis shows that this can be carried out efficiently with little loss of accuracy. © The Chartered Institution of Building Services Engineers 2001.