Research On Fatigue Driving Detection Method Based On Real Vehicle Data

The increase in car ownership has made the traffic safety situation more severe,and fatigue driving is one of the important incentives for traffic accidents.How to effectively identify and warn the driver’s fatigue state is of great significance to ensure the safety of the driver’s life and property and create a good road traffic environment.In view of the fact that high-speed road conditions are prone to fatigue driving,this paper mainly studies the driving behavior data under real vehicle conditions.The data has highly practical value owing to its non-invasive and unaffected by the driving enviroment features.This paper constructs a driving sample library through data preprocessing and evaluation,screens out the optimal feature parameter around driving behavior feature analysis,feature index extraction and feature parameter optimization,finally develops a detection method for fatigue level under real vehicle condition,which based on building up the identification model by using decision tree and neural network.The specific research contents are as follows:1.Relying on VBOX to collect platform data,build a database of fatigue driving samples.Firstly,to preprocess the driving behavior data under real vehicle conditions.The data with steering wheel angle greater than ± 20 ° and vehicle speed lower than 70 km / h are eliminated to ensure the effectiveness of subsequent extraction of fatigue characteristic indicators;Secondly,to eliminate the zero drift of steering wheel angle.That is to elect the steering wheel angle of a straight line to calculate the mean value,obtaining that the angle zero drift value is 5.46 °,then subtract this value from all angle data;Finally,to segment and evaluate the data.It is to obtain the marked driving behavior data set of 5 people,and to make statistics on the sample data of level 3 fatigue level of each person.2.To analyze driving behavior to characterize fatigue characteristics,and screen the optimal feature parameter set.The waveforms of steering wheel angle,yaw rate,vehicle speed,lateral acceleration and longitudinal acceleration under different fatigue levels are analyzed.And it is found that the waveforms mainly show changes in frequency and amplitude.According to this principle,46 statistical indicators are extracted from the driving behavior data,and the optimal characteristic parameter set representing the fatigue level is selected by one-way analysis of variance,which are the steering wheel angle standard deviation time window 16s(SWA-Std-16),steering wheel angle small amplitude frequency percentage(SWA-PSAF),vehicle speed mean time window 20s(Speed-M-20),yaw rate standard deviation time window 16s(Yaw Rate-Std-16)and lateral acceleration standard deviation time window 11(X＿Accel-Std-11)these five characteristic parameters.3.To study the fatigue driving detection algorithm and to build the driving state identification model.The decision tree and neural network are selected as the main algorithm for building the model,and the input is the optimal feature parameter set.Through the establishment of the identification fatigue model on the SPSS software and Tensorflow platform,under the real vehicle operating condition data,the accuracy of the decision tree in identifying the third-level fatigue is 74.26%,the accuracy of the RBF neural network in identifying the third-level fatigue is 74.16%,and LSTM neural network identification accuracy reached 78.65%.