Expressway Driving Behavior Recognition Based On GPS Data

Establishing expressway warning system is an important measure to improve expressway traffic safety condition.As the key technology of warning system the recognition of expressway driving behavior has become the research hotspot and attracted attentions from scholars all over the world.With the development of GPS and other positioning technologies,achieving expressway traffic monitoring and driving behavior warning will become a reality by exploiting GPS positioning data.Therefore,it is of great significance to recognize the expressway driving behavior based on GPS positioning technology.Firstly,we classified the driving behavior and conduct the real driving experiment.Through the analysis of the driving process on the expressway,the driving behavior are divided into four types: Free Driving,Car-Following Behavior,Lane Changing Behavior and Overtaking Behavior.Specifically,according to the speed and initiative of the vehicle,the car-following behavior is divided into Initiative Car-Following of High Speed,Passive Car-Following of Low Speed Initiative Car-Following of Low Speed and Passive Car-Following of Low Speed.Similarly,the lane changing behavior is classified into Lane Changing of Out and Lane Changing of Back based on the target lane.As for the overtaking behavior,Left Overtaking and Right Overtaking are defined according to the direction of overtaking.Besides,the driving behavior experiment was conducted on the Ring expressway of Harbin by making use of the high-precision GPS equipment.Through the data processing,the driving behavior characterization parameters are obtained.Then the characteristics of expressway driving behavior are analyzed.At the fist,we analyze the parameters describing the four types of car following behavior.These parameters include vehicle Speed,the acceleration rate,the following distance,the following headway,and distance between the front and rear vehicles,After that,the significance test is conducted and the result shows great differences between four kinds of driving behavior.As for the lane changing behavior,the parameters featuring it are also obtained by analyzing the trajectory curvature,vehicle speed and acceleration rate,lanechanging duration,and the headways with respect to the front and the rear vehicles.Then,we analyze the overtaking speed and distance and duration characteristics of vehicles choosing to overtake and a negative exponential function is adopted to fit the overtaking speed differences with the overtaking duration and the overtake distance respectively with theFinally,the driving behavior recognition model is constructed by using the Hidden Markov Model,which is tested and evaluated at the end.Based on the previous analysis and the principal component analysis,eight parameters characterizing driving behavior are selected.Then,the Hidden Markov Model is established for each driving behavior,in which the overtaking behavior model is actually a combination of OLC and BLC models.After setting the basic model parameters,we optimize,the length of time series,the number of hidden states,the Gauss mixture number and parameter combinations for each sub model through testing the model Therefore,the training of the model is completed and the driving behavior data is obtained..Finally,our model test result indicates that the driving behavior model established can be used to identify the driving behavior with a high accuracy rate.When using unclassified samples,the driving behavior recognition model can identify OLC and BLC behavior after the occurrence of these behaviors within 1.0 or 1.1 second.