Particle filters for location estimation in ubiquitous computing: A case study

Abstract

Location estimation is an important part of many ubiquitous computing systems. Particle niters are simulation-based probabilistic approximations which the robotics community has shown to be effective for tracking robots' positions. This paper presents a case study of applying particle filters to location estimation for ubiquitous computing. Using trace logs from a deployed multi-sensor location system, we show that particle filters can be as accurate as common deterministic algorithms. We also present performance results showing it is practical to run particle filters on devices ranging from high-end servers to handhelds. Finally, we discuss the general advantages of using probabilistic methods in location systems for ubiquitous computing, including the ability to fuse data from different sensor types and to provide probability distributions to higher-level services and applications. Based on this case study, we conclude that particle filters are a good choice to implement location estimation for ubiquitous computing. © Springer-Verlag 2004.