Sample-based Probabilistic Estimation for Indoor Positioning and Tracking Under Ranging Uncertainty
Yuan Yang · Universitätsbibliothek der FU Berlin Hochschulschriftenstelle u. Dokumentenserver · 2015
The critical problem in range-based indoor positioning is the severe ranging uncertainty, which typically resorts to the probabilistic perspective. Since there is no analytical solution to the nonlinear and non-Gaussian positioning problem, the research trends have moved towards exploring sample-based approximations in the probabilistic frame. However, the sample-based methods are generally inaccessible for application not only because of the high complexity, but also for the sampling difficulty and divergence. This thesis studies the sample-based probabilistic positioning to achieve the trade in performance (accuracy and robustness), cost (time and space complexity), and usability (in terms of the required number of samples and implementation difficulty). The work of this thesis covers both the theoretical and practical sides of sample-based probabilistic algorithms positioning. The proposed algorithms are tested by extensive simulations and real-world experiments on the Nanotron platform. Specifically, the contributions of this thesis can be summarized in the following: (a) Chapter 3.3 characterizes the indoor TOF ranging, i.e., the relationship of the ranging error to time, space, anchor connectivity and the ground truth of the ranging, etc. Then, intensive distribution fitting, hypothesis test, and model verification of different parametric models are carried out. (b) Via the imposition of the state constraints from the most recent observation, four constrained sampling methods are developed (see Chapter 4), with the advantages, including: 1) effectively reduces the sample size; 2) suppress sample degeneracy and impoverishment without resampling; 3) no requirement of keeping all samples’ properties in memory. (c) The NLOS mitigation proposed in Chapter 4 refines the measurement model as a positively biased and right-tail distribution. (d) Both the theoretical and practical anchor deployment are suggested in Chapter 4. (e) To address the estimation instability and sparsity problems, the one time-step smoothing methods are incorporated in the sample-based Bayesian estimation.