Improving the Heading Accuracy in Indoor Pedestrian Navigation Based on a Decision Tree and Kalman Filter

Abstract

In pedestrian inertial navigation, multi-sensor fusion is often used to obtain accurate heading estimates. As a widely distributed signal source, the geomagnetic field is convenient to provide sufficiently accurate heading angles. Unfortunately, there is a broad presence of artificial magnetic perturbations in indoor environments, leading to difficulties in geomagnetic correction. In this paper, by analyzing the spatial distribution model of the magnetic interference field on the geomagnetic field, two quantitative features have been found to be crucial in distinguishing normal magnetic data from anomalies. By leveraging these two features and the classification and regression tree (CART) algorithm, we trained a decision tree that is capable of extracting magnetic data from distorted measurements. Furthermore, this well-trained decision tree can be used as a reject gate in a Kalman filter. By combining the decision tree and Kalman filter, a high-precision indoor pedestrian navigation system based on a magnetically assisted inertial system is proposed. This system is then validated in a real indoor environment, and the results show that our system delivers state-of-the-art positioning performance. Compared to other baseline algorithms, an improvement of over 70% in the positioning accuracy is achieved.