Cloud tracking with optical flow for short-term solar forecasting
Year: 2012
Type: conference-paper
Abstract: A method for tracking and predicting cloud movement using ground based sky imagery is presented. Sequences of partial sky images, with each image taken one second apart with a size of 640 by 480 pixels, were processed to determine the time taken for clouds to reach a user defined region in the image or the Sun. The clouds were first identified by segmenting the image based on the difference between the blue and red colour channels, producing a binary detection image. Good features to track were then located in the image and tracked utilising the Lucas-Kanade method for optical flow. From the trajectory of the tracked features and the binary detection image, cloud signals were generated. The trajectory of the individual features were used to determine the risky cloud signals (signals that pass over the user defined region or Sun). Time to collision estimates were produced based on merging these risky cloud signals. Estimates of times up to 40 seconds were achieved with error in the estimate increasing when the estimated time is larger. The method presented has the potential for tracking clouds travelling in different directions and at different velocities.
Authors: Phillip Wood-Bradley, José Zapata, John Pye
Language: English
Subfield: Computer Vision and Pattern Recognition
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W2218681948
