The Improved Constraint Methods for Foot-Mounted PDR System

The Zero-velocity Update (ZUPT) aided Extended Kalman Filter (EKF) is commonly used in the classical INS-based foot-mounted PDR (Pedestrian Dead Reckoning) system. However, in the realistic test, the system still often suffers from drift, which is mainly caused by two reasons: failed detection of the stationary phase in the dynamic pedestrian gait and the heading drift which is a poorly observable variable of the ZUPT method. In this paper, in order to overcome these problems, three improved constraint algorithms have been proposed respectively for the detection of the stationary phase, constraint of heading drift and constraint of the height divergence. Firstly, for the problem of failed detection of the stationary phase, a novel stationary phase detection method is proposed which mainly by finding the minimum detector T in each gait cycle to determine the stationary phase, rather than totally based on threshold comparison principle in the traditional method. Comparing with the traditional method, the proposed method can detect the stationary phase of each gait cycle accurately under various pedestrian movements. Secondly, for the heading divergence problem, an improved method is proposed based on the existing HDE (Heuristic Drift Elimination) method, which uses the position error rather than heading error to restrain the trajectory divergence. Comparing with the traditional method, the proposed method can better constrain the heading to the correct angle. At last, for the problem of height divergence, an effective method has been proposed to determine the state of pedestrians by using the slope between adjacent one/several footsteps. At the same time, the slope of the plane and stairs is used to restrict the height divergence. In experiment, the proposed method can be more effectively to distinguish the pedestrian's state. Especially, when pedestrian walks on the stairs, the slope of the stairs can constrain the height divergence obviously.

Abstract licence: CC BY-NC-ND

A person counts once a day per paper. We keep no cookie and no address: a reader is a hash that changes every day. Any month under five reads shows no number.