Driving Behavior Analysis

The process of identification of the driver's driving state is the process of pattern recognition of driving behavior, but the use of pattern recognition method in the identification of driving state requires to combine with the specific characteristics of the driver's driving behavior. The data of driving behavior have characteristics of nonlinear and volatility. With the deepening of the driver fatigue, these two features will more apparent. Therefore, the study of pattern recognition algorithms suitable for driving behavior data is of great signification to improve the recognition rate.This paper takes the driver's driving behavior signals collected from driving simulation chamber as the object of study, such as brake, accelerator, steering wheel angle, cars off track distance from the center line. The use of Gaussian mixture model algorithm is studying the related algorithm in the process of pattern recognition of driving state, and analyzing the application of feasibility and effectiveness of Gaussian mixture model in the use of driving state recognition.Various data analysis and processing are compared. The main work of paper is as follows:1) Screening and analysis of raw dataThere are large the experimental data obtained from cockpit simulation experiments, and the experimental sites is circular experimental track, and the driver has different drivin behavior in different sections, and the data collected also has different characteristics. Driver's test time is at noon, this time period belonges of high incidence of fatigue accidents, the driver will appear several times short-time fatigue or close to fatigue state. Take these factors into account, select part of normal driving data and fatigue driving data to analyze and model.2) Create two-dimensional Gaussian mixture modelDrow on the two-dimensional Gaussian mixture model alogrithmand EM algorithm, take the steering angle and vehicle off track from the centerline as the input signals. The driving behavior signal(the steering angle and vehicle off track from the centerline) of three different drivers are trained and normal driving model of different driver's are established.3) Verify validity of the modelCompare the distribution of the contour of the model which established on the algorithm of Gaussian mixture model with the distribution of the original data to determine its effectiveness.4) Recognition rate of model The test data will be brought into the normal model estabilshed. The recognition rate of model has been determined by the percentage of the correct recognition account for the total test section.