Multi-View mmWave Radar Imaging with Few Measurements Based on Random Phase Shifting

High-resolution mmWave radar imaging plays an important role in applications such as autonomous driving. Beam-based imaging methods often require scanning the scene of interest with a sufficiently small angular stepsize to achieve high-resolution, thus leading to a large computational and storage burden. Compressive sensing (CS) is a promising strategy to reconstruct high-dimensional yet sparse signals from low-dimensional measurements with random sampling. Therefore in this paper, we conduct random space sampling by adding random phase shifts on the transmit antennas of a frequency modulated continuous wave (FMCW)-based radar system. We prove that the sensing model can be formulated as a CS problem and solved by Expectation-Maximization Gaussian-Mixture Approximate Message Passing (EMGMAMP)-based approaches. Simulation results show that the model has excellent imaging performance even with very few sensing measurements. To further improve the imaging quality, we consider a multi-view sensing scenario in which sensing results from different positions are fused by proper occlusion processing and coordinate transformation. Finally, appropriate evaluation metrics are proposed for target sensing results to validate the effectiveness of the proposed sensing model and algorithm.