Handheld SAR with Learning-Based Ego-Motion Estimation Using a Compact mmWave Sensor

Affordable millimeter-wave (mmWave) devices enabled radar-based computational sensing systems to be used as the primary modality in a wide range of applications from vital sign monitoring to surveillance. Such low-power and lightweight mmWave sensors have recently surged the development of portable handheld radar imaging systems operating using synthetic aperture radar (SAR) principles. These systems, however, are commonly coupled with external positioning devices which introduces hardware complexity, cost, and weight. Nonetheless, to the best of our knowledge, solely radar-based ego-motion estimation methods, common to automotive literature, have not been applied to handheld SAR imaging. In this paper, we aim to achieve ego-motion estimation for handheld SAR imaging using a compact sensor equipped with only 3 receiving antennas. Our work highlights the potential of this technique and addresses the limitations that arise by introducing a first prototype deep learning-based approach.