SquiggleMilli: Approximating SAR Imaging on Mobile Millimeter-Wave Devices
summary
System approximates SAR on hand-held 5G mmWave phones by combining non-linear motion compensation using multiple receiver antennas, standard compressed sensing to fill missing samples, density-based segmentation, and a conditional GAN trained on thousands of pairs of low-res mmWave and AR-camera ground-truth shapes. Implemented on TI mmWave radar with squiggle-collected poses, it reconstructs human-perceptible 2D shapes, predicts 3D features and class, and outperforms traditional SAR on occluded pistols and scissors.
AI-written summary · about
related talks
26mNext generation machine learning for mobile and embedded platforms
41mComputer Vision for Remote AR
20mDesigning a Narrowband Radar using GNU Radio and Software Defined Radio for Tomography....
19mSymmetriSense: Enabling Near-Surface Interactivity on Glossy Surfaces using a Single Commodity ...
21mRadarCat: Radar Categorization for Input & Interaction
30mPassive bistatic RADAR using spaceborne Sentinel1 non-cooperative source...
29mUSRP based X-band Digital Beam Forming Synthetic Aperture Imaging Radar
24mEnhancing Physical Perimeter Defense Using SDR