Contactless RF-based Sensing for Object Monitoring and Verification
Abstract:
RF (radio frequency)-based sensors provide significant advantages, such as detecting objects, measuring distances, and tracking movement without requiring direct line-of-sight. Their ability to penetrate materials like walls or packaging makes them ideal for use in complex environments such as healthcare, industrial automation, and logistics. The contactless nature of RF sensing enables remote monitoring in hazardous or inaccessible locations, reducing physical wear and enhancing long-term system reliability. This thesis explores the potential of contactless RF-based sensors by focusing on two key capabilities: monitoring small mechanical vibrations and reconstructing object shapes under non-line-of-sight (NLOS) conditions.
The first part of this thesis introduces PADrone, a system that leverages frequency-modulated continuous wave (FMCW) radar operating in the 5.8 GHz ISM band to detect drone abnormalities prior to flight. As drone delivery becomes increasingly adopted for autonomous logistics, ensuring system integrity is crucial. PADrone meets this need by identifying subtle vibration signatures in drone components during pre-flight inspection, achieving a fault detection accuracy of 97.5% and enhancing operational safety and reliability.
Similarly, the rapid growth of e-commerce, with billions of parcels delivered worldwide, has intensified the need for reliable package verification systems. To tackle the challenge of verifying items enclosed in packages—common in e-commerce and customs—we propose two NLOS reconstruction systems. PathScan, a 77 GHz mmWave radar system, addresses this challenge by verifying package contents without opening them. PathScan uses structured radar scanning with a mmWave sensor guided by a CNC (Computer Numerical Control) machine to perform multi-view Synthetic Aperture Radar (SAR)-inspired reconstruction. This allows it to estimate the bounding volume and shape of enclosed objects without unsealing the package. PathScan accurately reconstructs both dimensions and cross-sectional outlines, including for non-metallic items. Across 37 objects (25 metal, 12 non-metal), it achieves a verification accuracy of 97.9% with only 2.1% false positives. Comprehensive evaluations have demonstrated its effectiveness under various packaging types, item positions, and orientations.
To enhance practicality, we propose Scanify, a flexible system that eliminates mechanical constraints by co-locating a mmWave radar with a monocular camera. Leveraging COLMAP for structure-from-motion, Scanify estimates the radar’s path and fuses multi-view radar data to reconstruct 3D object geometry. While PathScan offers higher reconstruction accuracy through controlled, predefined motion paths, Scanify enables handheld, unconstrained scanning, trading some precision for deployment versatility. Together, these systems demonstrate the feasibility of contactless item verification in real-world, dynamic environments. We conduct a comprehensive evaluation of Scanify across diverse object shapes, scanning conditions, and movement speeds, demonstrating promising results.