Research On Driving Behavior And Driving Pattern Recognition Based On Mobile Intelligent Terminals

In recent years,Chinese automobile industry has developed vigorously,and the drivers’ unregulated driving behaviors have caused a lot of traffic safety problems.By using mobile intelligent terminals,it is convenient to obtain the driving data of vehicles,analyze the driver’s driving behavior with its pattern.In this paper,the method of driving behavior recognition based on mobile intelligent terminal is studied in order to identify the driving pattern.The main research work includes the following contents.Firstly,the acceleration sensor and gyroscope within mobile intelligent terminal are used to capture the acceleration and angular velocity information of the vehicle as our driving data.Before the driving behavior identification,the moving mean filter is selected to reduce the noise of driving data disturbed by vehicle jitter.Then,the endpoint detection algorithm popularly in the field of speech recognition is used to extract the switching points of different driving behavior segments in the driving data,thereby reducing the calculation amount of driving behavior recognition.Secondly,after analyzing the characteristics of different driving behaviors segments,22 dimension time domain features were extracted from the X and Y axis of acceleration sensor and the z-axis data of gyroscope.Considering the correlation between features,the principal component analysis(PCA)is used to reduce the feature dimension,and four features that best characterized the lateral and longitudinal driving behavior of the vehicle are obtained.Then k-means clustering algorithm is used to cluster the driving behavior data to obtain the sample distribution of the clustering results,after the calculation of DBI index,the optimal clustering number is confirmed to be 12,and FCM clustering algorithm is used to finely cluster the driving behavior data.Finally,the clustering centers of different driving behaviors are extracted and used as words in driving behavior dictionary.The word number of different driving behavior is calculated according to word frequency of driving data and its weighted histogram characteristics is obtained.Taking the shortage of the current mainstream topic models pLSA and LDA into account,this paper proposes an improved LDA model with time labels,namely the T-LDA model to identify driving patterns.The experimental results show that the improved model can effectively extract a series of continuous driving behavior characteristic of the driving data and improve the accuracy of driving pattern recognition.