Aggressive Driving Behavior Detection Based On Driving Status Data

Aggressive driving is one of the main factors affecting road traffic safety.Detection of aggressive driving behavior in vehicle driving has become a research hotspot in the field of intelligent transportation.In this context,based on the large data of driving,this paper studies the vehicle running status and proposes an aggressive driving behavior detection framework.The framework firstly uses Hank matrix to construct windows and change point detection algorithm to extract potential aggressive driving events from driving state data,then uses echo state network to extract driving event characteristics,and finally constructs two-level clustering model to cluster driving events.Through descriptive statistical analysis of clustering results,aggressive driving behavior can be detected.The main work of this paper is as follows:(1)In view of the influence of various factors on the data acquisition process,which leads to various problems in data quality,the data are pre-processed by filling missing values,deleting duplicate values and processing outliers,and the data with frequency anomalies and recording duplicates are screened and cleaned.Then,the change point detection of traffic state data is carried out.In order to consider the sequence samples in each time interval,a method of constructing a sliding time window using Hanker matrix is proposed.The density ratio change point detection algorithm based on RuLSIF is used to calculate the difference of adjacent time windows,and the change points of data points are obtained.The total change points of all features at the same time are accumulated.Number,using the change point to determine the rule to confirm the change point,and then extract driving events with potentially aggressive driving behaviors.(2)Echo State Network(ESN) is used to extract features of driving events with potentially aggressive driving behavior.The dynamic reserve pool is constructed by using the parameters of the network,and the driving event vector is input into the network training of the reserve pool.According to the mean square error between the actual output and the expected output of the ESN network training,the cross-validation method is used to select the optimal parameters,and the output connection weight matrix under the optimal parameters is extracted as the driving event characteristics.(3)A two-level clustering model based on the SOM algorithm and Mini Batch K-means algorithm are constructed.Firstly,the SOM clustering algorithm is used to cluster the driving event feature data in the first layer,create the initial cluster,get the number of clusters and cluster centers,and introduce U matrix to visualize the clustering results.Then,the first level clustering results are used as the initial input of Mini Batch K-means clustering algorithm,and the second level fine clustering is carried out.Through statistical analysis of clustered clusters of two-level clustering model,aggressive driving behavior was detected and clustering performance was evaluated by using contour coefficient.Finally,the proposed two-level clustering model is compared with the detection results based on DENCLUE clustering.The experimental results show that the proposed method can achieve excellent detection accuracy and execution efficiency.The aggressive driving behavior detection framework proposed in this paper is validated by using real data collected by the U.S.Department of Transportation.We extracted 24 vehicle driving status data.In the experimental ring of Inter Xeon 1.60 GHz CPU and 32.0 GB memory,we detect five kinds of aggressive driving behaviors: acceleration,acceleration,steep turning,steep acceleration turning and steep deceleration turning.The detection accuracy is 94.68%,and 600 driving incidents can be detected in 3 minutes.