Magnifisense

Inferring Device Interaction using Wrist-Worn Passive Magneto-Inductive Sensors

 
magnifisense_thumb.jpg

Ubicomp 2015: Edward Jay Wang, Tien-Jui Lee, Mayank Goel, Sidhant Gupta, Shwetak Patel


The different electronic devices we use on a daily basis produce distinct electromagnetic radiation due to differences in their underlying electrical components. We present MagnifiSense, a low-power wearable system that uses three passive magneto-inductive sensors and a minimal ADC setup to identify the device a person is operating. MagnifiSense achieves this by analyzing near-field electromagnetic radiation from common components such as the motors, rectifiers, and modulators. We conducted a staged, in-the-wild evaluation where an instrumented participant used a set of devices in a variety of settings in the home such as cooking and outdoors such as commuting in a vehicle. MagnifiSense achieves a classification accuracy of 82.6% using a model-agnostic classifier and 94.0% using a model-specific classifier. In a 24-hour naturalistic deployment, MagnifiSense correctly identified 25 of the total 29 events, while achieving a low false positive rate of 0.65% during 20.5 hours of non-activity.

devicetypes.PNG

After testing 33 types of devices, 12 of the most commonly found devices were chosen for evaluation to cover a diverse set of electronic components and contextual locations.

figureSignalProperty.png

EMI radiation patterns of commonly found electronic devices depend on the underlying electronic component.

resultsCM.PNG

The model-agnostic classifier accuracy averaged 83.9% across 12 classes. Accuracy was particularly low for the battery-operated motors such as shavers (67%), toothbrush (67%), and the car (46%). However, the out of bag accuracy from the random forest classifier produced an accuracy of 99.8%, which in our case is the model-specific accuracy. The out of bag accuracy, being an unbiased estimate of the test set accuracy, along with the RF classifier being highly unlikely to be over fit, suggests that there are potentially differences between brands of the same class that makes merging under the same class infeasible.