Simplified Pedestrian Tracking Filters with Positioning and Foot-Mounted Inertial Sensors

Henar Martin,Eduardo Metola,Ana M Bernardos,Juan A Besada,José R Casar

doi:10.1155/2014/850835

Abstract

Pedestrian tracking is one of the bases for many ubiquitous context-aware services, but it is still an open issue in indoor environments or when GPS estimations are not optimal. In this paper, we propose two novel different data fusion algorithms to track a pedestrian using current positioning technologies (i.e., GPS, received signal strength localization from Wi-Fi or Bluetooth networks, etc.) and low cost inertial sensors. In particular, the algorithms rely, respectively, on an extended Kalman filter (EKF) and a simplified complementary Kalman filter (KF). Both approaches have been tested with real data, showing clear accuracy improvement with respect to raw positioning data, with much reduced computational cost with respect to previous high performance solutions in literature. The fusion of both inputs is done in a loosely coupled way, so the system can adapt to the infrastructure that is available at a specific moment, delivering both outdoors and indoors solutions.

Highlights

The need to know where a person is dates back to centuries ago; in the last decades, traditional maps and compasses have been increasingly replaced by navigation assistance systems
Approaches such as the ones described in [15, 16] include an inertial navigation system that works in parallel with that complementary Kalman filter (KF)
With the same measurement rates and assumed percentage of zero velocity updates (ZVU) duration this processing would demand a computational load in the order of 475000 FLOPS

Summary

Introduction

The need to know where a person is dates back to centuries ago; in the last decades, traditional maps and compasses have been increasingly replaced by navigation assistance systems. Many existing solutions are based on Kalman filters (KF), which provide a reduced computational load with respect to particle filters Some of those works propose a complementary Kalman filter that estimates the errors of the position, velocity, acceleration, and the bias of the sensors, subtracting them from the states and measurements for the integration. Approaches such as the ones described in [15, 16] include an inertial navigation system that works in parallel with that complementary KF.

Pedestrian Movement Modeling

Measurement Modeling

Findings

Result and Discussion

Conclusions and Future Work