Total Variation Compressive Sensing for 3D Shape Estimation in Short-Range Imaging Radars

The problem of angular estimation for extended targets with a single snapshot in short-range imaging radar is considered. Two methods based on the framework of Total Variation Compressive Sensing, where sparsity is enforced in the signal’s gradient, are proposed. An independent azimuth-elevation estimation for each range cell is used in the first method, while a joint 3D range-azimuth-elevation optimization problem is proposed in the second one. The proposed algorithms have been verified on simulated data generated in a realistic manner using 3D CAD models of different cars, pedestrians, and bicycles. The results of the two methods have been compared with a conventional Fourier beamformer and the standard Compressive Sensing algorithm in terms of shape reconstruction. The Intersection over Union between the ground truth shape and the estimated shape with the four methods has been computed to evaluate the shape reconstruction capabilities. It is demonstrated that the presented methods outperform the other two methods, while other beamformers, such as MUSIC or ESPRIT, are not applicable in practice.