Due to differences in personality, driving skills and driving experience, drivers tend to show different driving characteristics. This paper studies the driving behavior characteristics though a real vehicle test. The differences among drivers analyzed by vehicle motion parameters, including acceleration and angular velocity. Kruskal-Wallis test results show that the acceleration and angular velocity of drivers are significantly different. To capture the time features, driving data is regarded as time series. In order to solve the problem of different time lengths, the dynamic warping time (DTW) method is adopted. DTW distance shows that the distance between the driver and himself is the smallest, and the distance between the driver and other drivers is all greater. It indicates that the driver has his personal driving habits. For further analyze the characteristics of drivers, the driving behavior data is converted from time domain to frequency domain by using the Fourier transform. And the differences between drivers in frequency are compared by calculating the correlation coefficient. The results show that the correlation coefficient of driver’s own acceleration is higher, and correlation coefficient of different drivers is lower. While both angular velocity correlation coefficient of driver’s own and different drivers are lower. It indicates that acceleration can better represent the driver characteristics. In order to capture the characteristics of driving data in frequency and time domain simultaneously, discrete wavelet transform (DWT) is used. To explore the performance of different parameters of individual characteristics. A driver identification model is established by Linear Discriminant Analysis (LDA). The result shows that wavelet energy entropy offers the best identification accuracy.

CICTP