A novel clustering strategy for fingerprinting-based localization system to reduce the searching time

Location estimation is essential to the success of location based services. Since GPS does not work well in indoor and the urban areas, several indoor localization systems have been proposed in the literature. Among these, the fingerprinting-based localization systems involving two phases: training phase and positioning phase, are used mostly. In the training phase, a radio map is constructed by collecting the received signal strength (RSS) measurements at a set of known training locations. In the positioning phase, the training location whose corresponding RSS pattern matches best with the currently observed RSS pattern is selected as the estimated location of the object. The positioning accuracy of such systems depends on the grain size of the training locations, i.e., better localization accuracy can be achieved with increasing number of training locations, which in turn, increases the comparison cost as well as the searching time in the positioning phase. Several clustering strategies have been proposed in the literature to reduce the searching time by grouping several training locations into a cluster and selecting the right cluster in the positioning phase followed by searching within the selected cluster to localize an object. However, selection of some false cluster degrades the positioning accuracy of the localization system. Thus, this paper aims at devising some novel clustering strategy that would reduce the searching time without compromising the positioning accuracy.