4D Millimeter-Wave Radar Point Cloud Processing via Global Optimization and Dynamic Enhancement

To address challenges in 4D millimeter-wave radar point clouds for unmanned ground systems (UGS), such as false alarms, elevation artifacts, sparsity, and under-utilization of velocity information, a collaborative optimization framework is proposed. It integrates point cloud optimization and dynamic target enhancement. Elevation artifacts are suppressed using an enhanced Shapiro-Wilk joint test with spatial entropy preservation, coupled with a cross-domain multimodal cascade optimization filter for progressive verify-cation and filtering. For dynamic targets, a dual-enhancement strategy generates directional perturbation noise along motion trajectories and applies curvature-adaptive gain modulation to improve geometric consistency. Experiments on the View-of-Delft (VoD) dataset exhibit a 65.07% anomaly reduction with intact spatial structure, increased GT box point densities, enhanced shape matching and refined trajectories precision. This process framework establishes an effective approach for sparse point cloud processing in UGS perception.