Persons tracking with Gaussian process joint particle filtering

This paper presents an approach to tracking persons using Gaussian Processes (GP) and Particle Filtering (PF). We used a binary switch sensor floor, which provides a natural and transparent way to build an indoor positioning and tracking system. However, it poses many challenges by producing nonlinear non-Gaussian measurements of true location. To solve these issues we present a novel algorithm. It uses PF for Bayesian tracking and data association combined with learned GP regression to correct estimates. Furthermore, the proposed algorithm, called Gaussian Process Joint Particle Filtering (GPJPF), handles multiple targets, where each particle models the targets' states jointly. To handle the data association problem and interaction between targets in close proximity, a Markov Random Fields (MRF) -based motion model was applied. Along with the GP model, it can be used directly as an additional factor when calculating the importance weights of particles. In comparison, the proposed method outperforms conventional Gaussian process and particle filtering methods.