Classification of Driving Modes Based on Driving Styles under Natural Environment

Classification for driving styles is a dominant factor of driving safety in natural scenes. Unluckily, the real-time and multi-dimensional characteristics are always ignored due to the fact that characteristics of vehicle motion parameters are merely focused on any traditional driving style classification. By contrast, the heterogeneities of driving mode characteristics are very remarkable among drivers with different driving styles, which could reflect drivers’ aggressiveness under the natural environment vividly. As such, a corresponding classification method was put forward based on driving modes. Firstly, classification is performed in view of velocity and driving directions. Sub-classification is then carried out in view of those factors such as target states and time headway to gain 18 driving modes. Then, the one-way analysis of variance (ANOVA) method is applied to select driving modes which have statistically significant differences between drivers with various styles. Finally, driving styles are classified utilizing the driving cycle multilayer perceptron (DCMLP) integrated with driving modes; also, corresponding training and verification are carried out in combination with driving styles which are calibrated as 3 classes (namely aggressive, common and conservative classes) based on driving data of 44 drivers. Our results indicate that the overall accuracy rate is up to 95.1% for our classification of driving modes whose accurate rates can be above 87%. Aggressive, common and conservative drivers prefer to relative approaching, target-tracking lane change and stable car following, and free lane changing, relative dispersion and free cruising conditions, respectively. The overall accuracy of DCMLP is up to 95.5% which grows by 25.6% compared to those of traditional numerical feature-based classification methods. Our classification method is coupled with traffic environment characteristic parameters so that it can be regarded as a new idea for the classification of driving styles.