Driving Behavior Analysis And Application Research Based On Trajectory

Analyzing the driver’s behavior characteristics and evaluaing the risk of driver’s behavior on the basis of the driving characteristics extracted from the driving trajectory data,which has broad application prospects in driver training,insurance industry,government traffic management and other fields.Traditional driver behavior analysis algorithms mostly use vehicle sensor data and driver’s physiological characteristics data,which is difficult to obtain and process.Some of the driving behavior analysis algorithms using trajectory data are subjective,and the analysis results are not objective;in addition.Additionally,the accuracy of the collected trajectory data is poor,which will have adverse effects on the result analysis.In recent years,with the continuous development of big data processing methods,the above problems have been gradually solved.This paper proposes a method of using high-precision trajectory data to analyze driving behavior.The main research work is as follows:(1)In view of the poor accuracy of the collected satellite positioning data,which can not meet the accuracy requirements of the existing research on the trajectory data,this paper proposes a method to improve the driving trajectory accuracy based on the Gated Recurrent Unit.This method makes full use of the strong nonlinear feature learning ability of the recurrent neural network on the time series data,adds the noise which obeys the bimodal two-dimensional Gaussian distribution to the highprecision trajectory,takes the position and velocity information as the input of the network,and takes the high-precision trajectory data as the label,continuously trains the proposed network until the proposed loss function converges.The simulation results show that the proposed method based on GRU network is better than the traditional method in noise reduction performance and speed.(2)In order to solve the problems of human subjective components and difficult data processing in current driving style analysis,this paper uses unsupervised learning K-Means++ clustering algorithm to classify trajectory data.Track data is filtered from the selected dataset,and driving characteristic parameters of each track are extracted.The dimension of each track is reduced by principal component analysis.The results of principal component analysis is the input of K-Means++algorithm to complete the clustering work.Finally,the silhouette coefficient method is used to analyze the clustering results,and compared with the traditional K-Means algorithm,it is found that the algorithm used in this paper has better clustering effect and less clustering time.(3)Based on the proposed trajectory accuracy improvement algorithm and driving behavior analysis algorithm,this paper designs a driving assistant system,which mainly realizes the functions of driving danger warning and driving behavior evaluation based on the real-time collected trajectory data.The system deployment is based on the "end management cloud" framework.After the real-time data collected by the vehicle terminal is improved,it is transmitted to the cloud computing center to calculate the real-time status information of the vehicle.The early warning module will monitor the safety of the vehicle in real time according to the calculation results,and take corresponding early warning measures when necessary.