Research On Drivers’ Fatigue Detection Method Based On Image Processing

At present, the traffic accidents have led to lots of economic loss and fatal casualty yearly, and the driver’s fatigue condition is one of the important causes of vicious traffic accidents. Thus, how to distinguish if the drivers are in the state of fatigue rapidly and accurately is of great significance to effectively decrease traffic accidents. To properly detect the condition of drivers’eyes is an effective way to study the state of drivers’fatigue degree because the opening degree of eyes reflects divers’ states of fatigue. By analyzing and studying the video images of complex luminous beams and changing gestures of the drivers who are actually driving on the road, this thesis has realized the accurate locating of drivers’ faces, further located the position of drivers’ eyes along with judging the states of eyes and finally come up with drivers’ fatigue conditions.The localization and tracking of the drivers’ faces under the complex conditions have firstly been achieved in this thesis. According to the feature that the colour of skin possesses excellent assembling character in the color space, this thesis has selected YCbCr color space as distributed statistical mapping room, built a two-dimensional Gause distributed mathematical model based on skin colour and utilized approximate elliptical match as a substitute of this model to achieve the purpose of rapid segregation of skin regions. In order to remove background interference, the method of marking connected domains as to remove several non-skin regions has been used, followed by the method of gray integral projection to preliminarily locate human faces and at last the method of centroid adjustment to eliminate the interference of the regions connected to human faces and realize the localization of human faces. This method which combines the three methods above has realized the accurate locating of drivers’ faces while the drivers are driving under the conditions of luminous beam changing, background interference, head turning and some other complicated conditions. Kalman filtering method has been used to realize the tracking of human faces on the basis of accurate human faces locating. These thesis employes the method combined with simplified generalized symmetry transformation and skin colour information to locate drivers’ yes in the complex luminous beams on the basis of human faces locating. Because of the small amount of computation and fast computing speed featured by gray image, the gray image has been used to locate eyes. Firstly, median filtering has been adopted to pre-process the framed image. And then, the complexity algorithm has been used to initially locate human eyes. In the areas of human eyes that have been initially located, the method which combines simplified generalized symmetry transformation and skin colour information has been adopted to achieve accurate locating of the drivers’ eyes. To decrease the amount of calculation and increase computational accuracy, the maximization of interclass variance threshold value segmentation method has been used to conduct binarization to the initial located areas of human eyes before applying simplified general symmetry transformation. And the value of generalized symmetry transformation is then calculated in the position that the result of binarization is zero (black). This method has realized the precise human eyes locating of the human face images in the condition of wearing glasses, complex luminous beams, slightly turning cheeks and neck wrying.On the basis of precisely locating human eyes, this thesis has used edge detection operator and the method of calculating the vertical height of eyes to judge the degree of eye closure and applied PERCLOS principle to determine if the driver is fatique and if we should alarm.MATLAB has been adopted in this thesis to develop the algorithm to realize the drivers’ faces locating, eyes locating and eyes’ condition detecting. Also, several related experiments have been made, the results of which have realized the anticipated goals and approved the validity of the algorithm that has been studied in this thesis.