Towards calibration-free geo-localization of stationary outdoor webcams
Original version
Sandnes, F. (2010). Towards calibration-free geo-localization of stationary outdoor webcams. I: E. Hjelmås (Red.), Norsk Informatikkonferanse. Trondheim: Tapir akademiske forlag. http://www.nik.no/2010/19-Sandnes.pdfAbstract
This study proposes two strategies for determining the approximate
geographical location of outdoor webcams based on time-series comprising
regularly sampled images. The strategies require an accurate account of
universal time and the date to be known, then the intensity of the images
are used to estimate the sunrise and sunset times. Given accurate sunrise
and sunset times a celestial model can be used to extract the latitude and
longitude of the webcam. Webcams vary in quality, dynamic pixel depth
and sensitivity to light. Two strategies are therefore proposed for avoiding
the need to perform calibration. The first technique involves normalizing
and noise-damping the image intensity measurements. The second technique
employs a self-normalizing brightness ratio. The brightness ratio is computed
from the overall brightness of the upper part of the image in relation to the
bottom part. During day the sky is much brighter than the ground, while
at night the sky is much darker than the ground if the ground is lit up.
Experiments demonstrate that the intensity normalization strategy is the most
robust and it is capable of determining the geographical location of webcams
with an accuracy of approximately 2 degrees.