Analysis And Evaluation Of Driving Styles And Maneuvers Based On Driving Operational Information Under The Highway Situation

As the number of cars increasing day by day,the bad driving behavior of drivers seriously affects the road traffic safety.In the current era of intelligent networked vehicles,monitoring driving behavior and even driving style effectively is significant on improving the safety of road traffic.At present,the off-line identification of driving style by empirically selected driving operation characteristics and traditional machine learning methods has limitations and incompleteness of evaluation dimension.Therefore,this paper analyzed the driving operation and behavior characteristics of drivers with different driving styles,and explored the deep learning identification method of driving style and the identification method of driving patterns.In order to explore the characteristics of driver’s driving operations and patterns under different driving styles,statistical analysis of driving operation data characteristics,driving pattern duration and frequency characteristics was employed.The analysis results of the driving operation characteristics show that drivers with different driving styles only have significant differences in vertical operation when driving forward,and the vehicle speed is maintained at different levels.For lane change pattern,significant differences were showed in horizontal operation.Drivers are more cautious when taking the left lane change against the right one.The analysis results of the driving mode characteristics show that free and straight travel is the main mode of daily driving of the driver,and the driver prefers restricted lane change to free lane change.In addition,the higher the driving risk,the more inclined the driver is to approach the front vehicle and follow up closely.However,due to driving safety or driving intentions,the driver would perform emergency braking or restricted lane changes.Aiming at the problem of driving style identification under high-speed conditions,this paper combined the driver’s instantaneous driving behavior and long-term driving effect to construct a driving operation diagram which represents driving styles.In addition,comparing with the support vector machine method,the convolutional neural network(CNN),long short-term memory(LSTM)network and pre-trained LSTM were supplemented to identify the driving style.The identification results showed that the proposed driving style identification system can effectively identify different driving styles,among which the convolutional neural network had the best effect.The identification accuracy on the test set could reach 98.5%.To solve the driving pattern recognition problem,the method of synchronous time window for expanding characteristics was applied based on the driving operational data.Then a combination method of the significant analysis and feature selection algorithm was taken for reducing the feature dimensions.Hidden Markov Model was established to determine the type of driving patterns using time series data.The identification results on the test set showed that the features selected by the significance analysis and forward selection algorithm expressed the characteristics of driving patterns the best.The identification accuracy was 82.3%.