Generating Synthetic Short-Range FMCW Range-Doppler Maps Using Generative Adversarial Networks and Deep Convolutional Autoencoders

In this paper, we discuss the usage of Generative Adversarial Networks (GANs) and Deep Convolutional Autoen-coders (CAE) for creating synthetic Range-Doppler (RD) maps of Frequency-Modulated Continuous-Wave (FMCW) radars for a short-range situation with moving objects, based on measured RD maps of pedestrians and cyclists. Instead of using regular mathematical functions or heavy radar simulations, we have used an Artificial Neural Network (ANN) model to generate new data. By using our synthetic data, we can automatically have ground-truth data without the need for manual labor; easily create large synthetic datasets; hardly use much computational power after training. To evaluate our method, we have trained a detector system with just synthetic data, and it was capable of detecting moving objects correctly, on actual Range-Doppler maps, 11.6% better than when using a small dataset.