What is it about?
Radar sensors can recognise people from the tiny chest movements caused by breathing and heartbeat, without cameras or microphones. Like most AI systems, however, they tend to confidently mistake a stranger for someone they already know. We developed MVGSig, a method that looks at the same radar signal in six different ways and combines these views with a graph-based neural network. When the views disagree, the system treats this as a sign that the person may be unknown. On public radar data from ten people, MVGSig cut the rate at which strangers were wrongly accepted by almost half.
Featured Image
Photo by Denis Nuțiu on Unsplash
Why is it important?
Contactless radar could make access control in homes, hospitals and vehicles more private than cameras. For that to be safe, the system must reliably reject people it has never seen. Our results show that combining several views of the signal, and letting stronger views guide weaker ones during training, makes radar identification much less likely to let an unknown person in, an important step towards trustworthy, privacy-preserving sensing.
Perspectives
This work grew out of my PhD at Auckland University of Technology. One finding surprised me: the signal views that are best at telling enrolled people apart are not the ones that are best at spotting strangers. That is why we combine all of them rather than picking the single "best" view.
Mohammad Hossein Shirazi
Auckland University of Technology
Read the Original
This page is a summary of: MVGSig: Multi-View Graph Fusion for Out-of-Distribution Detection in FMCW Radar-Based Person Identification, ACM Transactions on Internet of Things, November 2026, ACM (Association for Computing Machinery),
DOI: 10.1145/3862635.
You can read the full text:
Contributors
The following have contributed to this page
