RadarCat: Radar Categorization for Input & Interaction
summary
A 60 GHz FMCW radar on a chip (Google Soli 60 GHz FMCW chip) collects six‑antenna, eight‑channel returns; feature vectors (512-point channels + 21 hand features) feed a random forest classifier trained on 26 materials, 16 transparent samples and 10 body parts from six participants. Benchmarks show ~96% on a six‑object office set, high accuracy on transparent materials, and mixed results on some body parts (forearm/belly/calf confusion). Four working demos run in real time: object dictionary, painting/photo controls, body shortcuts and automatic refill.
AI-written summary · about
chapters (14)
- 0:00Introduction and overview
- 1:07Radar on a chip technology
- 3:13Object and material recognition demo
- 6:13Why radar-based recognition matters
- 7:37Comparison with other sensing methods
- 9:24How RadarCat system works
- 12:20Machine learning pipeline details
- 12:43Experiment results overview
- 13:51Transparent materials recognition
- 14:32Body parts detection results
- 15:37Potential applications
- 16:38Context-aware interaction demos
- 17:42Future work and limitations
- 18:11Q&A session
related talks
06m3D Point Cloud Generation with Millimeter-Wave Radar
06mSquiggleMilli: Approximating SAR Imaging on Mobile Millimeter-Wave Devices
40m"Introduction to Face Processing with Computer Vision|"
55mCore ML: Vision
14mDemo: Smart Embedded Vision with PolarFire® SoC FPGA
14mVeCare: Statistical Acoustic Sensing for Automotive In-Cabin Monitoring
19mSymmetriSense: Enabling Near-Surface Interactivity on Glossy Surfaces using a Single Commodity ...
12mCascade models in computer vision to boost accuracy and performance