- 860870.pdf (7M)
Tapir Akademisk Forlag
Most weather reports are either based on data from dedicated weather stations, satellite images, manual measurements or forecasts. In this paper a system that automatically generates weather reports using the contents on webcam images are proposed. There are thousands of openly available webcams on the Internet that provide images in real time. A webcam image can reveal much about the weather conditions at a particular site and this study demonstrates a strategy for automatically classifying a webcam scene into cloudy, partially cloudy, sunny, foggy and night. The system has been run for several months collecting 60 Gb of image data from webcams across the world. The reports are available through an interactive web-based interface. A selection of benchmark images was manually tagged to assess the accuracy of the weather classification which reached a success rate of 67.3%.
This paper was presented at the NIK-2011 conference; see http://www.nik.no/.
Permanent URL (for citation purposes)