Structure-coupled Variational Bayesian Method for Building Layout Reconstruction

This paper presents a structure-coupled sparse Bayesian learning method for building layout reconstruction using through-the-wall radar. We characterize the azimuth continuity of the wall and two-dimensional extensibility of corner into a hierarchical probabilistic model. Moreover, we employ a variational expectation maximization algorithm to perform Bayesian inference, which provides an internal building structure map. Meanwhile, a generalized approximate message passing algorithm is integrated to speed up operations by avoiding matrix inversion. Results based on real data validate that the proposed method can extract the walls and corners accurately, thus enabling building layout reconstruction completely.