Research On Method Of Driver Drowsiness Detection Based On Fusion Of Multi-face Clues

Recent years, the number of vehicles increased rapidly, which brings a large amount of convenience to people 's daily life and work, but the traffic problems are getting worse. According to the statistics of relevant department, the traffic problems caused by fatigue driving take a great proportion, so developing a system to real-time detect the drowsiness state of the driver has great practical value, whose foundation is drowsiness detecting algorithm, therefore, the thesis proposes a drowsiness detecting algorithm based on the driver 's face image.A detecting algorithm, which is based on the fusion of facial multi-parameters of drowsiness, is proposed, analyzing the state of the driver 's eyes and mouth, and the facial expression, and using the fuzzy system to reason the drowsiness state of the driver. Firstly, the driver 's face is localized by the detecting algorithm of cascaded Adaboost based on Haar features, and then the marginal information of the face is enhanced using Gabor filter to localize the eyes and the mouth precisely, and the state of the eyes is classified as being open and closed by a linear SVM classifier trained by an rotation-invariant LBP features of the eye image s; and the state of the mouth is determined according to the opening-area and the ratio of width-height of the binary image of the mouth; meanwhile the drowsy expressions of the driver are discriminated using multi-view distance metric learning algorithm. Finally, the state of the driver can be reasoned by the fuzzy system using four drowsiness coefficients calculated by the state of the eyes and the mouth, and the facial expression, and there are different forewarning measures to the reasoned results.The results of the experiment s show the accuracy of the state discrimination algorithms of the eyes and the mouth, of which the average accuracy of the recognition meet requirements of practical detecting system, however, for the discrimination of facial expressions, although the circumstance is complex and there are a lot of disturbances outside, the correction rate of the algorithm proposed in this thesis can still satisfy requirements of the practical application; the reasonability of the fuzzy system used to information fusion is verified in the simulation experiment as well. Finally, the reasonability and integrity of the whole algorithm are proved in the fifth chapter of this thesis.