A Driving Fingerprint Map Method of Driving Characteristic Representation for Driver Identification

Safety and comfortability are important indicators to evaluate the performance of driver assistant systems (ADAS) and intelligent vehicle (IV) systems. Driver identification or driving characteristic learning is needed to achieve the comfortability function for ADAS or IV systems. The effectiveness of driver identification or driving characteristic learning is directly affected by driving characteristic representation method. This paper develops a new concept of the driving fingerprint map to represent driving characteristics. First, the driving scenes are classified using the ensemble learning method. Then, the feature selection method known as conditional mutual information maximization is used to select the representative features for describing driving characteristics. Finally, the sliding time window is applied to generate the driving fingerprint map based on the selected driving features. To verify the performance of the proposed driving fingerprint map method, a real vehicle experiment is conducted to obtain the test dataset. The driver identification results that used the original driving features and the proposed driving fingerprint map based on the deep convolutional neural network, support vector machine, and extreme learning machine are compared. The results show that the driving fingerprint map method can effectively describe driving characteristics.