Indoor localization and tracking have received considerable attention in recent years. They serve as an enabling technology that makes numerous context-aware services and applications such as people tracking, personalized information delivery, medicine and health care possible. Many localization/tracking methods have been proposed in the literature, and they can be categorized mainly into two classes: propagation model-based methods and empirical model-based methods. As the complex indoor radio characteristics are often difficult to be reflected in a propagation model, empirical model-based methods generally produce better estimates. Moreover, tracking algorithms are generally superior over non-tracking algorithms for localization since they incorporate motion constraints....[ Read more ]

Indoor localization and tracking have received considerable attention in recent years. They serve as an enabling technology that makes numerous context-aware services and applications such as people tracking, personalized information delivery, medicine and health care possible. Many localization/tracking methods have been proposed in the literature, and they can be categorized mainly into two classes: propagation model-based methods and empirical model-based methods. As the complex indoor radio characteristics are often difficult to be reflected in a propagation model, empirical model-based methods generally produce better estimates. Moreover, tracking algorithms are generally superior over non-tracking algorithms for localization since they incorporate motion constraints.

Empirical model-based localization methods, however, usually require a costly manual calibration effort. In this thesis, we focus on the design of accurate empirical model-based indoor WLAN tracking algorithms with reduced calibration effort. Our algorithms implicitly make use of the inherent structure of indoor environments and motion constraints, and utilize elegant machine learning methods, leading to improved estimation accuracy. A simple algorithm is proposed to dynamically adapt the radio map using the online measurements. In addition, an algorithm based on manifold learning is developed to further reduce the calibration effort by leveraging on unlabeled user traces which can be easily and cheaply collected. We demonstrate the excellent accuracy, reduced calibration effort and effectiveness of the proposed algorithms for tracking mobile users through field experiments in indoor WLAN environments. Compared to the conventional algorithms, our proposed algorithms can reduce the number of training data per survey points by a factor of 14, and the number of survey points by a factor of 2 or 3, while retaining high estimation accuracy.