| Popis: |
Nowadays, information handling is among the biggest scientific challenges. Information is freely offered in great amounts almost everywhere, through internet and other easily accessible sources, constituting an enormous pool of available data. The process of exploiting available data is not an easy task, since it involves not only finding the proper means to do so, but it also needs to be done in reasonable time. With the continuously increasing number of satellite images, geo-data is progressively increasing both in quantity and in resolution, thus, there is a need of faster and more robust techniques of processing them. Exploiting geo-information is of paramount importance, as it comprises the source of many useful applications such as mapping, property bordering, area discovering, and land use, as well as other geospatial applications. While the presented techniques can be implemented for a variety of big data problems, our study apply them in map update, proposing a methodology that successfully tackles the big data problem by taking advantage of the processing power of many units simultaneously. The main goal of this work is to initiate distributed and parallel techniques in mapping systems by implementing efficient map updating algorithms based on road network extraction methodology on satellite/aerial images. In particular, the proposed algorithms are used to update digital maps, exploiting the distributed and parallel systems’ capabilities while applying various single-machine techniques to greatly increase the overall performance. The performance results confirm our initial assumption that a parallel and distributed system can significantly reduce the processing time of big data images, without any loss of output quality, having the necessary robustness of a market-ready product. More specifically, our study shows that even though the base algorithm (without any optimizations) is used, by adding more processing nodes the computation time can even be reduced to one third of the single machine processing time. |
