Abstract

While GPS has become the de facto standard as an outdoor positioning infrastructure, researchers are still actively working on the problem of indoor positioning. This is an exciting area with many practical applications from indoor navigation to location-based advertising.

There are currently three key problems that challenge the adoption and widespread use of indoor positioning systems. 1) non-line-of-sight signal propagation due to the cluttered indoor environment, which corrupts radio distance measurements, and leads to significant position errors; 2) sparsity of reference points, which results in large uncertainty in position estimation; and 3) large spatial variation in the accuracy of location sensors, which is difficult to measure empirically.

In this talk, I will report on recent research carried out in the Oxford Sensor Networks Group to address these challenges. First, I will present two distinct approaches to addressing the non-line-of-sight (NLOS) problem, an unsupervised approach, based on the theory of compressed sensing, and a supervised learning approach based on features of the signal distribution. Second, I will show how it is possible to overcome the problem of infrastructure sparsity by exploiting encounters (wireless contacts) between moving nodes. Finally, I will present a novel Expectation Maximization algorithm that can be used to learn the accuracy of different location sensors in different parts of a building, without having to actually measure it empirically.