Bayesian filters for indoor localization using wireless sensor networks

In indoor and urban canyon environments, where the Global Navigation Satellite System (GNSS) line-of-sight signals are very weak, accurate localization using GNSS alone is very challenging. So, it becomes necessary to combine GNSS technology with other wireless systems for proper localization. Recently, various forms of Bayesian filters have been used for combining the sensor information in order to estimate the location of a moving receiver. However the performance of most of these filters relies on the accuracy of the assumed probabilistic model of the system. In this paper, we show how the performance of these filters vary by applying them to various applications with different probabilistic models. We mainly focus on a new type of sequential Monte Carlo (MC) filter, called the cost reference particle filter, and show that this filter is more robust compared to the other filters as it does not make any assumption about the statistical distribution of the noise. Apart from the analysis of results of synthetic experiments, we also present a detailed analysis of the results obtained in a real world application where the trajectory of a robot has been tracked by integrating the measurements obtained using a set of ZigBee and Ultra-Wide Band (UWB) sensors into the filtering algorithms.