Northwestern Team Unveils 92.5% Accurate Gesture-Sensing Smartwatch With 8 Electrodes
Updated
Updated · Hackster.io · Aug 11
Northwestern Team Unveils 92.5% Accurate Gesture-Sensing Smartwatch With 8 Electrodes
1 articles · Updated · Hackster.io · Aug 11
Summary
Northwestern researchers reported EITWatch, a smartwatch prototype that recognizes hand gestures by sending tiny electrical currents through the wrist and reading tissue changes instead of using cameras or bulky bands.
Eight gold-plated electrodes in a 31 mm ring under a 40 mm watch case feed impedance data at 48 Hz, while a new multi-depth scanning method probes different tissue depths to capture muscle and tendon motion.
12 participants helped test two gesture sets, and the system reached 92.5% accuracy for six larger hand poses and 91.5% for five micro-gestures plus rest within a single session.
That approach beat conventional adjacent-electrode measurements by 15.1 percentage points for larger gestures and 10.4 points for smaller ones, but accuracy fell to 73.2% after users re-wore the watch two days later.
Cross-user performance was weaker without personalized training, leaving the team to pursue representation and contrastive learning to make gesture recognition stable across wearings and different people.