In this paper we propose a robust, track before detect particle filtering algorithm for tracking the direction of arrival (DOA) of multiple echoes impinging on a sonar array. We also provide a real data application of our filter for underwater 3 dimensional (3-D) scene reconstruction, i.e. bathymetry. Shallow water environments generally involve multi-path propagation channels. Coupled with impulsive, heavy-tailed backscattered signals it generates spurious estimates. To counteract this situation, particle filtering enables taking into account the impulsive nature of signals and regularizing the DOA trajectories by means of state-space description. For a specific double target tracking scenario, an automatic target detection and initialization scheme is presented. The state-space description incorporates and merges the available prior information on the imaged scene. The derived filter is shown to provide better performances when compared to classical array processing methods for the considered remote sensing application.