Robust fingerprinting-based localization using directed graphical models

In this paper, we propose a robust fingerprinting-based localization using directed graphical model. To overcome the influence caused by the jitter of received signal strength (RSS), the location of one mobile node can be estimated by fusing both the current matching result and the previous location estimation, using the Bayesian graphical model (BGM). Then, the localization problem is cast as a maximum a posteriori (MAP) estimator, which is also proved to coincide to maximum likelihood (ML) estimator. However, the initialized MAP estimator can be hardly solved with incomplete statistical property concerning the random vectors. To this end, we propose an adaptive smoothing algorithm (ASA) to attain the suboptimal solution of the original problem. Finally, the experimental results show that the proposed algorithm obtains a significant performance gain.