Driver Fatigue Detection Algorithm Based On Eye States

Driver fatigue has become one of the most important causes of traffic accidents. In order to reduce the incidence of traffic accidents, enhance traffic safety, protect life and property safety, Driver fatigue detection system research has great significance.(1)Illumination processing method based on image processing technology has been studied. Histogram equalization has good treatment effect for the too bright or too dark image background but the detail treatment effect is very poor. Retinex has good effect to restore image detail.According to the feature of histogram equalization and Retinex.Fusion method is proposed based on wavelet transform. The corresponding experiments are made by Yale B face database. The comparison of the experiments showed that the fusion method can weaken the impact of the light on the image, properly preserve the details, enhance the image feature information and improve the clarity.(2)Face detection method is based on skin color model. Color distribution statistics use the YCbCr color space as the mapping space. However, the traditional YCbCr is susceptible to illumination changes. Color space is processed by non-linear. The establishment of a mathematical model is based on the color of the Gaussian.But when the background color is similar to the complexion, the situation will cause large errors. Therefore, we use the geometry of the face to proceed further testing on the basis of the similarity of the face. Simulation results show that the YCbCr color space can be more adaptable to illumination changes and provide a much higher accuracy rate.(3)A secondary human eye location method based on image processing technology has been studied. Integral projection function can extract better features, but accuracy is not high. Differential projection function may reflect different gradation conversion rate. Calculus projection is proposed by combination of integral and differential projection in the first step. After first step, we use Hough transform detect pupil round because Hough transform has anti-jamming features. Experiments by MATLAB platform show that the use of secondary human eye location method based on calculus projection function and Hough transform can effectively improve the adaptability and accuracy, and have better eye positioning result in some interfering substances (such as glasses).(4)In the eye-tracking process, we use Kalman filter predict the position where the eye will appear in the next frame. After estimated position has been confirmed, we search for the eye positioning by the template matching within a set range. Simulation results show that the combination of Kalman filter and template matching can adapted to changes in illumination and face rotation circumstances.(5)PERCLOS Algorithm is used to test driving fatigue. But when human eye has not been closed for a long time, the algorithm effect is poor. Fatigue can be judged by Blink frequency, but the accuracy is not high. When blink frequency is too low, it reflects driver's state is in trance. Therefore, we combine PERCLOS algorithm and blink frequency to determine the driver's fatigue. Human eye pupil hole circle which has been detected is used to calculated percentage of the human eye. Through the percentage, we make judgments about state of the human eye. Results show that the frequency of blinking combined with PERCLOS that can enhance the accuracy of assessment fatigue.Driver fatigue detection algorithm based on eye states presented in this paper has strong anti-interference, adaptability and high accuracy. It is worth with certain theoretical significance and practical value.