Indoor Mobility Models For Wireless Positioning
Permanent address of the item is
Indoor positioning technology has become increasingly popular in both business and research worlds. Several technologies have been developed so far and some of them are in commercial use now. However, due to personal privacy issues and the complexity of indoor environment, the data regarding the human mobility patterns are insufficient. The study of synthetic human mobility models is an important issue, which is expected to shed new light into a myriad of Location Based Services and location-aware communications. Finding and testing synthetic models about human mobility is an important step ahead and this constitutes the main focus of this thesis. In addition, we also address the issue of indoor positioning via WiFi received signal strength under various mobility patterns, generated synthetically through a simulator built within this thesis. The thesis starts with a review of four popular synthetic human mobility models which is followed by presenting a new model proposed in this work and denoted as Hybrid Model. Based on the suitability of the models for indoor positioning, the Random Direction Mobility Model and the newly proposed Hybrid models were chosen for further testing as human mobility models with WiFi-based fingerprinting. We show in detail the indoor scenarios characterization and accordingly we present the classical path loss model. Then, an indoor positioning simulator including mobility models is built and an alternative method of evaluating Access Points (APs) deployment is introduced. In order to explore the positioning accuracy of the above two models, a fingerprinting algorithm with Bayesian combining is applied. The results are shown in terms of Root Mean Square Error (RMSE) distance error. Finally we conclude that a Hybrid Model has a better positioning accuracy than a Random Direction Mobility Model and that neither of the two models is essentially affected by the velocity range or by the variation of the starting point. We also show how the noise variance affects the positioning results.