A Preliminary Study on Synthesizing mmWave Radar Range-Doppler Maps under Occlusion

Deep Learning (DL) has recently expanded the potential of radar technology. Integrating neural networks with radar automatically extracts features from radar images, enhancing the accuracy of target detection. However, these models require substantial volumes of labeled radar data, and the labeling process is resource-intensive, time-consuming, and costly due to its reliance on manual labor or complex simulation. Generative Adversarial Network (GAN) has emerged as a promising alternative for generating synthetic radar data. Existing studies have primarily focused on using GAN to generate single-target radar data, ignoring mutual effects, especially occlusion caused by multiple targets. In this work, GAN is trained with simulated radar data under different occlusion degrees and used to synthesize Frequency Modulated Continuous Wave (FMCW) radar Range-Doppler maps (RDMs) based on given input. The evaluation results show that the agreement rate between simulation and synthetic data is more than 96% in low and severe occlusion cases, highlighting the potential of synthetic data for millimeter Wave (mmWave) radar RDMs under occlusion.