6 1. IntroductionAfter evaluating aspects relating to the nature of the propagation models, the othermajor part of the evaluation will focus on the performance of the localization al-gorithms. The three described algorithms, the HMM, the PF and the LMSE, willuse the resulting propagation models from the first part of the evaluation to solvethe localization problem on an evaluation corpus consisting of sequences of locationannotated RSSI readings.These sequences represent measurements from eight differently defined paths of var-ious complexity. For example, the most demanding one with respect to the localiza-tion problem is a path upward through the stairways over three levels of the building.For each of two different Android devices, an Iconia tablet and a Nexus smartphone,160 sample paths were taken. This leads to an evaluated distance of about 8000 mduring the real world evaluation.But before analysing the results on these real world measurements, a synthetic set ofmeasurements with 320 samples over the eight paths will be employed under differentnoise conditions. This idealized environment will help to evaluate the differencesbetween the HMM, the PF and the LMSE with respect to their algorithmic nature.Afterwards the evaluation will be finalized by using results of the radio propagationevaluation and the experience from the synthetic evaluation for interpreting theobservations that are made in the real world evaluation. It will be seen, that thepromising results from the synthetic evaluation are not directly mappable to thelocalization in natural environments.Furthermore, the localization algorithms are employed on the results of the radiopropagation evaluation that relate to the granularity of the 3D geometry. It will beseen, how complex the scene needs to be modelled to derive a propagation modelfrom the PHOTON raytracer that leads to acceptable localization error rates. Inthe last part of the evaluation, it will additionally be investigated how well theframework generalizes over more than one device. A propagation model that hasonly seen measurements from one device will be evaluated on the other. The resultsof this experiment seem promising.1.5 OutlineThe structure of this thesis is given as follows: After this introduction, the conceptsused in the framework and needed for understanding the localization algorithmsare explained in the background chapter 2 which is followed by the related workchapter 3. In chapter 3, comparable approaches to the localization problem foundin the literature will be investigated. By building on the foundations lain in thebackground chapter, the design chapter 4 is structured. There, all major componentsof the framework and their interactions are described comprehensively. An overviewof the implementation is given in chapter 5 which is finally followed by the thoroughevaluation of the system presented in chapter 6. In the last chapter 7 conclusionswill be drawn and an outlook into further research activities will be given.