Recognition of Fatigue Driving Based on Frequency Features of Wearable Device Data

Abstract

Fatigue driving is a primary reason of traffic accidents. Recognition of driver's fatigue state, prompting and supervision in time will effectively reduce traffic accidents. At present, fatigue driving detection methods mainly focus on physiological detection and image recognition. Physiological detection requires more sensors on the tester, which has a great impact on the driver. Image recognition is greatly influenced by environment. Given the growing popularity of wearable smart watches with the capability to detect human hand movements, this paper studies the potential to recognize fatigue driving based on steering operation by using a wearable smart watch. The sensor data used includes acceleration and angular velocity data related to drivers' operation behavior under different states. To eliminate the effect of gravitational acceleration on the data values of acceleration sensor, the coordinate system of acceleration data is transformed to the world's coordinate system. The main advantages of smart watches are that there are many kinds of sensors, low cost and low power consumption. The frequency domain features are obtained by Fourier transform of the data collected by the sensors of Smart Watch, and the feature dimension is reduced to 10 dimensions by principal component analysis. Finally, the recognition model of fatigue driving based on support vector machine(SVM) is established. The results show that the proposed method recognizes the drivers' fatigue or normal state more effectively than others and its accuracy can reach 82.6%.