Unlocking health insights: Estimating advanced walking metrics with smartwatches
Google Research
Researchers developed a deep learning model using temporal convolutional networks to estimate gait metrics from smartwatch accelerometer and gyroscope data, enabling health analysis from wrist-worn devices. The model was validated on 246 participants across approximately 70,000 walking segments using a lab-grade reference system, achieving Pearson correlation coefficients greater than 0.80 for most metrics with performance comparable to smartphone-based methods despite using half the training data. Smartwatches can now provide continuous, accessible gait analysis outside clinical settings for early disease detection and fall risk assessment without requiring precise phone placement.
Why it matters
Health & Bioscience
