Real-time detection of aurora in ground-based all-sky images

The increasing number of ground-based auroral imagers has created new challenges for data management and processing. Moreover, there is a clear demand of real-time data products for both scientific and public outreach purposes. We are currently developing computer vision methods in order to provide quantitative information of auroral image contents. The first step in automated image analysis is to detect whether a given image actually contains auroras. In particular, we are interested in processing the stream of thumbnail images from the 20 all-sky imagers of the THEMIS (Time History of Events and Macroscale Interactions during Substorms) ground-based segment. The existing network technology does not make it possible to transfer full resolution data from the remote stations to the data distribution centre. Furthermore, there is not enough time to process the images locally at the stations either. This means that real-time detection has to be based on 32 by 32 pixel gray-scale thumbnail images. We compare simple and computationally light methods for detecting the auroras in thumbnails. We also contrast the detection results to those of processing colour thumbnail images, in which the additional colour information can be used to separate background from auroras more efficiently. The presented methods can be generalised to handle most auroral all-sky images including the thumbnails in the Global Auroral Imaging Access virtual observatory (GAIA-VxO, http://gaia-vxo.org).