The Sdudy On The Discriminating Method Of Drunk Driving Based On The Driving Behavior Analysis

With the continuous development of social economy, and with the continued growth of vehicle conservation, traffic accidents occurred more and more frequently. Road traffic accident has become one of the main causes of human casualties, about 50%-60% of the traffic accidents were related to drunk driving. Therefore, it is of great significance to identify drunk driving accurately. The driving experiment was conducted in a driving simulator. The feature parameters that could represent drunk driving significantly were extracted through the analysis of driving behaviors in drunk state and normal state, and the driver's state was identified based on the extracted feature parameters. The main contents studied are as follows:Firstly, the experiment was designed to collect data based on the driving simulator. The present research at home and abroad was studied. On the basis of summarizing the predecessors' research, the driving experiment was conducted in a driving simulator. A total of 25 drivers' driving behavior data in different driving conditions were collected in the experiment. In the end, a sample database was established.Secondly, the effect mechanism of drunk driving on driving behavior was analyzed, and the feature was selected. Statistical methods were used to compare and analyze the mean and standard deviation of driving behavior in this thesis. It has been clear about the influence mechanism of drunk on the speed, acceleration, steering wheel angle, depth of the brake pedal, depth of the accelerator pedal and distance to center of lane. Ultimately, the steering wheel angle was extracted as the feature parameter for drunk driving detection.Thirdly, the sliding data window was used to extract feature. To reduce the feature loss through the statistic of mean and standard deviation by overall length of the data, this thesis used sliding data window to calculate the mean and standard deviation sequences of steering wheel angle to construct feature parameter respectively. The influence of different data window length on feature extraction and the effectiveness of the different data window to extract features were studied. The results show that sliding window data can be used to extract the feature parameter that can represent the drunk driving clearly.Fourthly, the approximate entropy and sample entropy were used to extract feature. First, the connection between parameters and the two entropy values of steering wheel angle were studied. The model of compute optimal entropy values was set. Second, the feature parameters were constructed based on the optimal entropy values and the performance of ApEn and SampEn as discriminate feature was examined by the ROC curve. The results show that the performance of approximate entropy of steering wheel angle as the discriminant feature was better than the sample entropy.Finally, the drunk driving detection model was established based on driving behavior. The KNN and SVM were used to establish the drunk driving detection model based on a single feature parameter and the weighted fusion multi feature parameters respectively. To compare with the recognition accuracy and efficiency of detection model, the results show that the SVM recognition effect is better than that of KNN; weighted fusion can improve the recognition accuracy of detection model. The SVM takes more time to search for the optimal parameters, and its efficiency is lower than the KNN.