This Is AuburnElectronic Theses and Dissertations

Pedestrian Navigation using Particle Filtering and a priori Building Maps

Date

2019-05-16

Author

Ray, Tanner

Type of Degree

Master's Thesis

Department

Mechanical Engineering

Abstract

This thesis presents a new particle filter (PF) weight update method that improves the performance of indoor positioning systems. In standalone inertial pedestrian-dead-reckoning (PDR) systems, the position error grows with time due to the inertial measurement unit's (IMU) sensor errors. Often external measurements from GPS or radio networks (e.g. wireless local area network (WLAN), ultra-wide-band (UWB), Bluetooth low energy (BLE) etc.) are used to restrict the error growth. External measurements from infrastructure-based systems have inherent high costs and deployment time, thus they are not easily implemented. The presented work focuses on the development of a standalone wearable navigation system that does not depend on physical infrastructure. In order to constrain error growth without external measurements, other techniques have been developed that utilize building map information as a measurement. One method uses the building to provide a heading measurement to reduce the drift in the heading solution. This is based upon the behavior that pedestrians typically walk straight when walking in building corridors. Another method constrains the error based upon the knowledge that pedestrian motion is limited by building floorplans, (e.g. walls, floors, and other features). This technique uses PF estimation to fuse standalone PDR with map measurements to perform accurate pedestrian localization. These techniques along with the current PDR techniques and underlying algorithms are discussed in detail. Lastly, this work presents a comparison of PFs that utilize different particle propagation and weight update methods for indoor positioning systems. A new type of weight update is also introduced that provides more accurate localization. The performance of the new weight update method is proven with a performance evaluation that includes both simulated and experimental data. The results of this and a summary are provided.