ABSTRACT
RF sensing in wireless communication networks is a novel approach for motion detection, but it faces challenges in accurately localising motion which is crucial for confinement in lighting control use cases. A probabilistic model enables motion localisation through sensor fusion. However, in probabilistic models the posterior estimations do not scale well with large networks as likelihoods of all possible system states need to be computed. It will be demonstrated that variational Bayesian techniques offer attractive approximations to the posteriors where the approximations require computational resources that scale with the number of nodes. This method is of general interest to large networks as it models nonlocal effects through localised updates.
