Recursive Bayesian Estimation using A Topological Map for Indoor Position Tracking

Target positioning and tracking become a very challenging topic for indoor applications. To improve the accuracy of indoor position tracking, it is useful to acquire map descriptions of indoor environments, with two major paradigms: accurate metric map (Mmap) and topological map (Tmap). Research has shown the efficiency of using Tmap for positioning, but often poses difficulty to explicitly represent environments. This paper proposes a recursive Bayesian filter incorporating Tmap and TOA (Time-of-Arrival) sensor ranging measurements for meter-level localization, namely T-loc. Constraining Bayesian recursion by both Tmap and ranging measurements not only substantially reduces the number of state samples, but also bounds the estimation error against non-line-of-sight (NLOS) ranging errors. Three Bayesian filters are tested in both simulations and real-world experiments, taking different target trajectories in a large-scale indoor scenario. Results show that T-loc outperforms generic particle filters and achieves an average localization error about 1 meter, indicating significant improvements compared to approaches without Tmap.