Step-Based Attitude Update (Sbupt) Technique For Pedestrian Dead Reckoning (Pdr) Using Handheld Devices

Pedestrian navigation using smartphones is a challenging problem that usually takes place in Global Navigation Satellite Systems (GNSS) deprived environments such as indoor and urban canyon. Such navigation techniques rely heavily on inertial measurements using Dead Reckoning (DR) techniques. The problem arises from two factors, the first being the use of low-cost Inertial Measurement Units (IMUs) integrated into smartphones that suffer from high deterministic and stochastic errors, and the second is due to the unconstrained motion of the sensing platform - the smartphone - with respect to the user. Pedestrian Dead Reckoning (PDR) is a special form of Dead Reckoning (DR) that utilizes information from human gait cycle to further enhance the performance over regular DR, but still remains unreliable due to the accumulation of errors from the integration of measurements over time. This paper proposes a novel attitude correction technique called Step-Based Attitude Update (SBUPT) that exploits information from the natural human motion mode. SBUPT uses the window of accelerations recorded during a step taken by the user to estimate the tilting angles of the smartphone, under the assumption that a user reaches a near constant velocity during regular motion mode. SBUPT uses a Least Squares (LS) estimation based on the covariance of the accelerations within the step and their average to accurately estimate the gravity components in the final body frame at the end of the step. SBUPT provides a solution to two common problems in attitude estimation, the first being accumulated errors from previous estimations, and the second is the window size selection for averaging-based alignment techniques. Correct estimation of the tilt angles is of great importance as it affects the linear acceleration and gravity separation, where small errors in gravity separation from the accelerations lead to high errors in step length and heading estimation. Tests show a root mean square error in tilt angles estimation below one degree in cases of smartphone compassing use cases, and below 2.5 degrees in case of a free-moving device.