SmartDetect: Safe Driving by Detecting Steering-Wheel Handling With a Single Smartwatch

Abstract

Holding the steering wheel with both hands is essential for safe driving. This article proposes a novel approach using only one off-the-shelf smartwatch to determine whether the driver is holding the steering wheel with both hands. Two classification models, namely, the individual and universal models, are proposed. An individual model focuses on a particular driver, while the universal model is applicable to all drivers. Both models extract vibration features from the watch's accelerometer signals using the Hilbert-Huang transform and classify the signal pattern by using support vector machines with a radial basis function kernel. Data samples were collected from 35 drivers. The universal model can achieve an accuracy of 98.51% for the hand on which a smartwatch is worn and 90.29% for the hand on which the smartwatch is not worn; the individual model achieves a higher accuracy of 99.21% for the hand on which a smartwatch is worn and 97.18% for the hand on which the smartwatch is not worn.