Research on Pedestrian Heading Estimation Based on Multiple Carrying Position Constraints of Smartphones

A pedestrian heading estimation method with multiple carrying position constraints is proposed, aiming at the heading inconsistency between smartphones and pedestrians due to various postures in the pedestrian inertial navigation system based on smartphones. After establishing the information fusion model of the attitude obtained by gyroscope, gravity sensor and magnetometer, an extended Kalman filter is given to improve the heading accuracy. Support Vector Machine classification assisted by Principal Component Analysis is used to accurately classify three common carrying positions, which are holding in front of the chest, swinging in hands and making calls. In the holding mode, the average of heading of the phone in one step is used as the pedestrian heading. Principal Component Analysis is also applied to extracting the main direction of X and Y axis acceleration under the navigation system in the swinging mode. The declinational angle is calculated before and after the carrying position switching to estimate the pedestrian heading in the calling mode and switching process. Experimental results show that the proposed method can effectively classify carrying positions during walking. The average heading error is less than 4.1°, and the offset error of the end point is 2.13% in the total distance of 120 m, which meets the precision requirements of the pedestrian navigation.