Improving data fusion in personal positioning systems for outdoor environments

A fault detection and correction methodology for personal positioning systems for outdoor environments is presented. We demonstrate its successful use in a system consisting of a global positioning system receiver and an inertial measurement unit. Localization is based on the dead reckoning algorithm. In order to obtain more reliable information from data fusion, which is carried out with Kalman filtering, the proposed methodology involves: (1) evaluation of the information provided by the sensors and (2) adaptability of the filtering. By carefully analyzing these factors we accomplish fault detection in different sources of information and in filtering. This allows us to apply corrections whenever the system requires it. Hence, our methodology consists of two stages. In the first stage, the evaluation is conducted. We apply the principles of causal diagnosis using possibility theory by defining states for normal behavior and for fault states. When a fault occurs, corrective measures are applied according to empirical knowledge. In the second stage, the consistency test of the filtering is performed. If this is inconsistent, principles of adaptive Kalman filtering are applied, which means the process and measurement noise matrices are tuned. Our results indicate a reasonable improvement of the trajectory obtained. At the same time, we can achieve consistent filtering, to obtain a more robust system and reliable information.