Concealed Object Detection with 3D Radar and DNN-Based Feature Extraction

Abstract

In the first part of the presentation we will discuss physics-inspired DNN architectures for leveraging mm-wave phased arrays to extract 4D radar features that can be used to classify a moving object or a static object from a moving phased array radar system. We will share how the concept was developed with the progress in both the system level and available compute, allowing a system integrator the possibility to optimize costs and power requirements. We will discuss tradeoffs with state-of-art MIMO radars and how we believe the DNN can go beyond the limits of conventional micro-doppler approaches. In the second part of the presentation, we will present our multi-modal DNN architecture, where both infrared and radar are being used to unlock capabilities that cannot be achieved by each of them separately, both in terms of accuracy and speed. We will share our feasibility study on leveraging how we used this DNN architecture to serve two different use cases: the first is an advanced human-computer-interaction and concealed object detection. This study shows the value in combining the two modalities compared with the results of each modality individually.