A KNOWLEDGE-BASED KALMAN FILTER FOR AN INTELLIGENT PEDESTRIAN NAVIGATION SYSTEM

Abstract

Continuous and reliable position determination is very important in any navigation application. Therefore a combination and integration of different location techniques and positioning sensors is required. In most navigation applications this integration is performed using a Kalman filter approach. In this paper a new approach which makes use of knowledge-based systems for preprocessing the sensor observations is presented. In the preprocessing step the quality and reliability of the sensor observations is tested and gross errors and outliers are detected and eliminated. Furthermore the preprocessing step is used to determine the weightings of the sensor observations in the stochastic model of the following central Kalman filter. The weightings of the sensor observations can then be adjusted in the filter depending on their availability and quality. This approach is developed in a research project at our University for a pedestrian navigation and guidance service. In this project different location techniques such as GNSS and indoor positioning are combined with dead reckoning sensors (e.g. digital compass for heading determination, accelerometers for measurement of travelled distance, barometric pressure sensor for altitude determination) for continuous position determination of a pedestrian user. The project takes a user case into account, i.e., the navigation and guidance of visitors of our university to certain offices and persons. Selected results of field tests using different sensors are also presented in the paper. From the tests it could be seen that such a service can achieve a high accuracy and reliability for continuous position determination of a pedestrian user. It can also be expected that the performance of the system can be increased using the new intelligent knowledge-based Kalman filter approach for the integration of all available sensor observations.