Optimization of fusion algorithm for hybrid pedestrian localization and navigation

Abstract

Hybrid pedestrian localization based on multiple data sources is becoming more and more popular. Nevertheless, accurate and reliable pedestrian localization is still a challenge due mainly to their unpredictable movement. For some applications such as interactive museum guidance unpredictable pedestrian movement is a major obstacle to accurate localization. In this paper we introduce a novel fusion algorithm using best-neighbor rating. The algorithm reduces the accumulated error originating from unreliable sensor measurements and increases the efficiency by only evaluating the nearby cells of the last estimated position. Experimental results show that a mean error of less than 1.5 M is achievable in real-world scenarios.