Research On Driver Gaze Zone Based On Head And Eye Features

With the rapid development of vehicle entertainment devices and communication systems,the driver's attention gets more and more interference in the driving process,resulting in increasing traffic accidents.In order to solve this problem,this paper focus on the research of driver gaze zone during driving.The existing gaze estimation system in the laboratory environment uses relatively simple lighting conditions,and the performance behavior is quite different from the real driving behavior.This paper proposes a driver sight area estimation method that combines head and eye features,and the effectiveness of the system is tested by real driving image.The main research contents of this paper are as follows:In order to extract the features of the driver's head and eyes,this paper first presents the method of locating driver's facial features:AdaBoost algorithm is used to locate the driver's face area.Aiming at the problem that more than one faces may be detected in the cab image when multiple people are sitting in the car,this paper determines the driver's face image by the relative position relationship between the face image center and the headrest of the main driver's seat;In this paper,ERT algorithm is used to locate the driver's facial feature points,and the effectiveness of the feature point location algorithm is verified through real driving images.In this paper,the driver's head features are used to estimate the area of vision at the moment of closing eyes:Firstly,the common area of the face center point is obtained by statistical analysis.According to the relative position of the center point in this area,the height and driving posture of the driver could be estimated;In this paper,instead of the traditional post algorithm to estimate the driver's head posture,we use the method of deep convolution network to eliminate the influence of the camera's internal parameters on the experiment,and evaluate the prediction effect of the network through the relevant public data set.In order to get the driver's precise gaze zone,the relevant eye features are extracted:Firstly,the driver's eye region image is obtained by facial feature points;then the driver's eye state is detected by SVM algorithm.If the eye is in closed state,the system will not extract the driver's eye feature;the initial position of the driver's pupil is obtained by the method of sliding template and image integral projection curve,and the accurate position of the driver's pupil is obtained by LBF cascade method;after getting the exact location of the driver's pupil,we first extract the driver's monocular feature through the relative position of the pupil in the eye region image,and then extract the driver's binocular feature through clustering algorithm.In order to verify the accuracy of the proposed algorithm,this paper establishes the gaze zone estimation data set in the real driving environment.The final comparison experiment proves that the driver's gaze zone estimation system proposed in this paper has achieved a high accuracy in the common area,which can meet the driver sight estimation needs under complex natural lighting conditions.