Study On Driving Fatigue Detection Based On Multi-Features

With the rapid development of our country’s economy,the sales of various types of cars have steadily increased,and people’s driving time has become more longer than before.Long-time driving will inevitably lead to fatigue driving accidents emerge in endlessly.Fatigue detection is the most effective method to reduce fatigue driving because of the characteristics of easy popularization,tracking detection,and strong real-time performance.At present,single-feature fatigue detection methods have great limitations for great effects from environment and individual differences.Therefore,it is great practical significant to develop an effective multi-feature fatigue detection method for ensuring road safety.The method of information fusion from various features of image signals and EEG signals was used to detect fatigue features at the Decision-Level to build a fatigue detection model,and it was shown better accuracy,stronger stability and higher reliability.The specific process is as follows:1)EEG(Electroencephalogram)data and image data were obtained in fatigue and non-fatigue driving conditions of simulating driving.The data set including 652 fatigue samples and 446 non-fatigue samples was established by the pre-process methods of eliminating invalid data,segmentation and synchronization,and label.2)After the original EEG signal from the simulating driving is filtered from 0.5 to 32 Hz,and EEG signals for detecting fatigue were gotten by removing the artifacts.δ,θ,α and βrhythms were extracted through discrete wavelet transform,and then their relative wavelet energies were calculated.The wavelet energies of α/β,(δ+θ)/(α+β),δ and α were used as the input of SVM to classify fatigue.It was found that these wavelet energies have a good detection effect as the EEG fatigue feature.3)After pre-processing including grayscale,median filtering,and histogram equalization,the face was quickly and accurately located by the AdaBoost algorithm based on MITEx face dataset training.and the detailed features of human eyes were extracted by the 194-point ASM model trained on the base of the HELEN 194 database.PERCLOSE value and blinking frequency were thought as the image fatigue detection features.4)After comparing the effect of two different fusion models of the comprehensive feature and the independent feature,the comprehensive feature data fusion model was chosen as the detection model because of its better detection effect.Facing to the traditional D-S evidence theory problem of a poor effect on the treatment of contradictory evidence,the improved D-S evidence theory was utilized as the decision level fusion method in this paper.In order to establish a fatigue detection model,SVM in soft classification output was used as the Mass function source of D-S evidence theory of improvement to achieve data fusion at the Decision-Level.After testing,the single EEG-feature fatigue detection accuracy rate is 85.5%,the single image-feature detection accuracy rate is 69.8%,and the detection accuracy rate based on the fatigue detection model is 91.3%.A multi-feature fatigue detection method that combines EEG fatigue features and image fatigue features was studied in this paper.Compared with a single EEG detection method or a single image detection method,it shown higher detection accuracy and the improving overall performance.