Data Driven Parameter Optimization Algorithm for RPM Based 3-D Radar Imaging

Microwave, millimeter wave imaging radar is indefensible sensor for self-driving system because of its applicability even for optically blurred vision, such as in dusty air, thick smog or fog. As a promising radar imaging method, the range points migration (RPM) has been developed, which offers more accurate target shape compared with a general synthetic aperture radar (SAR) imagery. However, its reconstruction accuracy largely depends on several parameters, which are usually determined in an empirical way. To address with the above problem, this paper introduces the Gaussian mixture model (GMM) for the RPM algorithm, where the parameter is automatically optimized by the expectation maximization (EM) algorithm. Furthermore, the large aperture data are appropriately processed without significant increase of computational cost, by introducing the iterative RPM processing and new weight function. The results from finite difference time domain (FDTD) based numerical test, introducing a realistic automobile model, verify that our proposed method achieves an automatic parameter optimization feature in data integration scheme.