1 of 10

Magic Room: Device-free Sensing of People

using batteryless tags and machine learning

Haoyan Zeng

​

​

Advisor: Prof. Bletsas

(Week 2)

2 of 10

Modelling

• The room is divided into grid pixels.

• Each pixel is a candidate human location.

• Each antenna-tag-frequency link gives one RF measurement.

• A human changes multiple RF paths through reflection, blocking, and scattering.

• Each propoagation path has own contribution

​

3 of 10

4 of 10

5 of 10

Solve the livnear inverse problem

6 of 10

7 of 10

8 of 10

9 of 10

10 of 10

From Model-Based DFL to ML-Based Sensing

​

Limitations of current methods

  • Fusion and clustering assume the number of targets is known
  • Close multi-person cases are difficult
  • Linear models struggle with real RF interactions
  • Ellipse-based voting is uncertain

​

Possible ML direction

  • Learn the mapping:RFID signal pattern → people count → location / zone

​