Pedestrian Dead Reckoning System Using Foot Kinematic Constraint and Multiple Virtual Track Technique

Abstract

The main objective of this dissertation is to enhance the performance of the pedestrian dead reckoning (PDR) system based on gait constraints. Two algorithms are proposed in this dissertation such as [predicted velocity update using kinematic constraint and shoe modeling] and [heading drift estimation based on multiple virtual track using dominant direction]. First, a PDR system that uses the velocity predicted by kinematic constraint conditions and ellipsoidal shoe assumption as a measurement value at the contact phase is proposed. Unlike the conventional PDR system that uses the zero velocity update in the stance where the shoe is completely stopped, the proposed PDR system uses the predict velocity at the contact phase where the shoe touches the ground. The predicted velocity is defined as a function of the lever-arm vector through kinematic constraints, and the relationship between the attitude of the shoe and the lever-arm vector is established through the ellipsoid modeling and the ellipsoidal parameter calibration by measuring the foot pressure distribution. We propose the PDR system suitable for various movements through filter design considering vibration and flexibility of shoe by detecting contact phase with ground by using acceleration predicted through kinematic constraint. The performance of the algorithm is verified by experiments considering various actions. Next, a heading drift estimation method using the dominant direction and multiple virtual paths is proposed for indoor navigation. Generally, PDR system using the dominant direction of the building can be used only when it is a building composed of narrow corridors or when a pedestrian walks along the corridor. In this study, we propose a method to utilize the dominant direction of the building by using the information of the previous time instead of judging the usability of the dominant direction of the building only at one point through the multiple virtual paths, and the position estimation performance is improved stably in various indoor environments.