Feature Detector: a support system for tracking satellite detected dynamic and permanent features

Abstract

Even with today's technologies many tasks relies on humans to be completed correctly. Engineers must monitor steps in large chains of operations, and verify the results before the next process is allowed to continue. In many such systems, a lot of useful data passes by without ever been stored for efficient future usage. Even though some operations must be verified by an experienced human eye, many, if not all, could benefit from computed assistance. When processing satellite imagery there are a lot of steps involved before there is a final product. This thesis examines one specific part of the process, how to determine if a feature observed is permanent or not. A self learning geographical information system implementation that can determine the state of specific features will be presented. The system is capable of filtering out permanent installations from vessel traffic in highly dense areas. Further we'll see that with such a system at the core, other useful functionality can easily be extended on top of it. Such functionality could be tracking of vessels, oil spills or ice floes, the latter two which have been implemented. With such a system at hand, the day to day tasks of engineers monitoring satellite observations can be made easier and less error prone. In addition such a historical view of the data can help with improving existing services as well as those still under development.