Study On Monitoring Driver Distraction Based On Binocular Vision

Driving safety is a hot issue all the time in the development of automobile industry.The statistics show that driver distraction is the main cause of vehicle near crash and mostroad traffic accidents. Therefore, it is of great significance for preventing traffic accidentsand reducing economic losses to detect driver’s attention states in time and take appropriatemeasures.On the basis of machine vision theory, the paper used two cameras and a near-infraredilluminator which had not disturbance to driver’s eyes to detect driver’s attention states. Inorder to eliminate the influence of light factors, the corresponding wavelength of IR filterwas adopted. This system was suited to the day and night driving environment. The mainresearch is listed as follows.1. Calibration of the binocular vision system is the foundation of stereo matching andthree-dimensional reconstruction. At first, binocular vision system was built and calibratedto detect driver’s attention states. The method of Zhang’s camera calibration was used tocalibrate each camera with plane-chessboard as the calibration reference. Afterwards,calibration of the binocular vision system was completed.2. The preprocessing and positioning of infrared face images are the basis ofextracting driver’s eyes feature information. Firstly, the Gaussian Filters algorithm wasapplied to smoothly preprocessing infrared face images. Secondly, the Otsu segmentationalgorithm was adopted to detect driver’s face region. Finally, the driver’s face region wasextracted using the vertical and horizontal projection algorithm.3. Driver’s eyes feature extraction is an important basis of the gaze directionestimation. Based on the positioned infrared face region, it was divided again to obtaineyes rough location. Then eyes were positioned precisely by the region labeling algorithm,on the basis of which, positions of the pupil and Purkinje spot were detected to estimateaccurate driver’s gaze direction. The fixed threshold segmentation algorithm was used tosegment eyes region again. Then the pupil position was located by the morphologicalfiltering processing and least squares ellipse fitting algorithm. At last, the Purkinje spotposition was acquired through Harris Corner Detection.4. Driver attention states were monitored according to eyes feature information. The corresponding feature points were obtained by stereo matching of pupil and Purkinje spot.Then combining with internal and external parameters matrix of the binocular visionsystem, three-dimensional space coordinates were acquired through three-dimensionalreconstruction. Driver’s gaze direction could be estimated by the three-dimensional methodbased on reflection from the pupil center and cornea. Above all, driver’s attention statescould be determined using the evaluation model.The software system and part of the algorithm which are used for monitoring driver’sattention states have been developed based on Visual C++2008. In summary, experimentswith infrared face images of different drivers were carried out to obtain driver’s attentionstates information, which lays the foundation for further research in future.