Research On Driver Vigilance Detection Based On Face Information Fusion

With the rapid development of economy,more than 300 million motor vehicles are in possession,which brings much convenience to the society.However,it also causes huge losses resulting from a large number of traffic accidents.Relevant studies and reports show that the decline of driver vigilance is the main causes of traffic accidents.Therefore,research on driver vigilance detection is a great of significance for improving road safety.Vigilance includes two main factors: attention level and fatigue state.However,usually only one factor is considered in traditional vigilance studies.In order to comprehensively detect driver's vigilance,based on method of driver face information fusion,this paper focuses on the key issues in location of face feature points,tracking of feature points,and construction of vigilance detection.The specific research works are as follows:1.Location of driver face feature points.Aiming at the problem that simple feature point location methods lack accuracy and robustness,this paper proposed a method based on modified active shape model(ASM).Firstly,a face ASM containing 26 feature points is established,and then face structure constraint was introduced into the average of synthetic exact filter(ASEF)to restrain unreasonable output,and further enhanced its robustness by rotating the filters.Then the improved ASEF was used to modify the face ASM during matching.Finally,experiment proves that the proposed method can effectively improve the accuracy of feature points loation.2.Tracking of driver face feature point.To deal with the problem that it is diffiult to detect sight direction,since the feature points location method can not adapt to large change of head posture,this paper proposed a feature points tracking method based on adaptive ASEF correlation filtering.Firstly,by introducing multi-template adaptation and floating strategy of learning rate into the ASEF correlation filter tracking algorithm,it improved the problem that fixed learning rate can not adapt to different target variation.Then,using the relative position change between the three feature points of double pupil and nose tip to adjust the scale of the target.Finally,experiment indicates that the porposed tracking method is effective.3.Construction of driver vigilance detection model.In view of the incomplete problem of tranditional driver vigilance detection,this paper designed some metheds to calculate each part of face information.Firstly,based on triangle sight model,distance of eyelid midpoints and distance of lip midpoints,the vigilance state parameters,such as sight direction,percentage of eyelid closure over the pupil over time(PERCLOS)and mouth opening degree were calculated.Then,the feature vector was obtained by cascade fusing those parameters,and finnaly,the driver vigilance detection model based on support vector machine(SVM)was constructed.Finally,experiment shows that the accuracy of the model reaches 93.8%,and the average time consumed per frame is 49.12 ms,which indicates that those proposed method can effectively detect the state of driver vigilance.