Heading Estimation for Multimode Pedestrian Dead Reckoning

The flexible carrying mode of smartphone brings challenge for Pedestrian Dead Reckoning (PDR) especially for heading estimation with build-in sensors. This paper focuses on POCKET mode and SWING mode and analyzes the correlation between smartphone's rotational motion and pedestrian's walking cycle states and walking direction. Based on the analysis, we propose to use the rotation vector sensor data of smartphone within one walking step to estimate the pedestrian's heading. For POCKET mode, heading is estimated by an improved rotational approach (IRA). A jitter detection algorithm is proposed to extract leg flexion interval. Stable walking direction without 180° ambiguity is obtained from the averaged rotation axis. For SWING mode, a single-point (SP) method is proposed. Heading is estimated from the direction of smartphone's y-axis when it is closest to the horizontal plane. The algorithms are validated with data collected by HUAWEI mate 10 smartphone. The RMS errors are less than 4.37° and 3.38° for POCKET and SWING mode respectively. Superior to previous heading estimation algorithms, our method can converge within one single walking step for both carrying modes without 180° ambiguity.