Research On Driver Fatigue Detection Method Based On Physiological Signals

Fatigue driving is one of the important causes of traffic accidents.How to accurately detect the drivers' fatigue status has become a hot topic for domestic and foreign researchers.The existing fatigue detection methods also have many shortcomings,such as low accuracy,high cost,poor popularity and driver comfort.Therefore,depth and detailed studying the driver fatigue detection techniques and methods,combining with the means of statistics,signal analysis and machine learning to develop high-precision fatigue detection technology become one of the important design goal of fatigue driving detection system.Physiological signal is an important detection method of human fatigue status.In this paper,we mainly detect the drive's fatigue through the human physiological signal.Through analyzing the advantages and disadvantages of fatigue detection based on Linear Classifier?SVC and MLP Classifier,we put forward two new methods,such as the feature-optimized random forest algorithm and the adaptive random forest algorithm to detect the driver's fatigue status.These new methods realized the purpose of improving the driver's fatigue detection accuracy.The main works of this thesis are as follows:According to the past fatigue driving detection system exist data interference and data collection expensive,this thesis adopt low cost and non-invasive physiological parameters,such as skin electrical conductivity,pulse oxygen content and respiratory signals to detect the driver's fatigue status and then use Hilbert transform to filter the collected data to achieve the goal of reducing fatigue driving detection cost and data interference.Aiming at the problem that the driver's fatigue detection accuracy is not very high,a method of fatigue detection based on feature-optimized random forest algorithm is proposed.By analyzing the relationship between the intensity and the correlation of random forest algorithm's base classifier and using the chi-square test,this thesis optimizes the feature selection of random forest algorithm and realized the goal of improving driver's fatigue detection accuracy.Experiments show that the proposed algorithm has improved the accuracy of SVC and MLP Classifier by 35.2%and 4.3%.Aiming at the problem that the scale of driver's fatigue samples is limited and the different influence of each sample on the detection results,a method of driver's fatigue detection based on quasi-adaptive random forest algorithm is proposed.By combining the advantages of the random forest algorithm's randomness characteristics and the Adaboost algorithm's adaptive characteristics,the thesis make the training samples and the base classifier adaptively adjusted and realized the goal of improving driver's fatigue detection robustness under the condition of ensuring the accuracy.Experiments show that the algorithm is 38%and 6.5%higher than SVC and MLP Classifier accuracy algorithms.In order to make the fatigue detection results more objective and authenticity,the experiment adopt the human physiological data in real driving environment and use parameters such as accuracy,mean square error,ROC curve,accuracy rate and recall rate to evaluate it.The thesis realized the goal of improving driver's fatigue detection accuracy in the case where there is interference in the sample set and the sample collection is small.