Estimating buildings layouts using exterior radar measurements is a challenging task involving the electromagnetic modeling, many unknown parameters, and limited number of sensors. We propose using the jump-diffusion algorithm as a powerful stochastic tool that can be used to determine the number of walls, estimate their unknown positions and other parameters. We improve the convergence rate of the jump-diffusion algorithm by developing an iterative procedure that first finds low- resolution estimates, which are then used to initiate our more accurate estimation. Our efficient usage of the available frequency bandwidth, improves the computational speed that otherwise would be hampered by the forward electromagnetic modeling.