Comparative study of sensor fusion methods for hybrid positioning systems

Abstract

Particle filter and Viterbi tracking algorithm are two representative sensor fusion methods to estimate a location of a mobile device by using various signals such as Wi-Fi, Bluetooth, magnetic fields and sensors like 3-axis accelerometer and gyroscope in indoor environments. This paper measures and compares the performance of the two sensor fusion methods on two paths in the first floor of KI building, KAIST. One path is a random walk and another is a rectangle path circumventing the area. On the rectangle path, the two methods showed a similar performance to each other, approximately 2m average error distance. On the other hand, on the random walk path, the Viterbi tracking algorithm significantly outperformed the particle filter in accuracy. This indicates that the choice of sensor fusion method influences a little to positioning accuracy when pedestrian dead reckoning (PDR) is working properly. However, when the PDR is not properly working, the choice of sensor fusion method greatly influences to positioning accuracy. In our test, the Viterbi tracking algorithm revealed to be much more resilient to PDR errors than particle filters.