CarOSense: Car Occupancy Sensing with the Ultra-Wideband Keyless Infrastructure
summary
CarOSense reuses the car’s UWB keyless infrastructure to sense per-seat occupancy by turning Channel Impulse Responses into a 4D tensor and feeding them to a deep multi-task CNN. The system instruments the cabin with 8 UWB nodes (Decawave DW1000 + STM32), aligns first-path taps, and trains a MaskMIMO multi-task mask model on data from 9 participants at 10 locations. It reports 94.6% leave-one-out accuracy and 87% under 10 unseen stress scenarios, runs on laptop, Google Coral and Raspberry Pi using TensorFlow Lite, and meets real-time constraints.
AI-written summary · about
related talks
13mAttraCar: Multisensory In-Car VR with Thermal, Airflow, and Motion Feedback through Built-In Vehi...
25mMulti-Vehicle Map Fusion Using GNU Radio
22mA Fully Immersive Driving Simulator for Autonomous Vehicles
13mInnovation with Wearables: Futurespective and Use Cases in Industry
10mTowards Context-aware Automatic Haptic Effect Generation for Home Theatre Environments
09mLightning Talk: Streaming and Processing Edge Vision Data in Real Time
41mKEYNOTE: Image processing, AI, and autonomous systems
45mRoom scale VR tracking with OpenHMD