A Study Of Driver’s Gaze Zone Estimation Algorithm

Recently, due to the increasing number of vehicles and consequently rapid growing of drivers, traffic accidents take place more frequently. As a research shown, most of the accidents are caused by driver’s inattention, some of which can be avoided by tracking the gaze direction. Gaze tracking technology is playing a vital role in studying driver’s attention, detecting fatigue and improving driver assistance safety. So it is attractive for domestic and foreign researchers.Even though some traditional gaze tracking systems have worked well in some imagined simple driving environments, they cannot be used in real driving since the disturbing factors such as highly variable illumination, large range of head movement and screen calibration which cannot aiming manually.To solve the problem of driver’s gaze zone estimation, this paper presents a new automatic calibration algorithm for gaze zone tracking. The method uses a single camera to capture driver’s face image, using face detection algorithm to get the accurate position and determining the center point of driver’s face as the feature point. Then, according to the driving environment and behavior, it utilizes side mirrors, rear-view mirror, dashboard and different areas of windshield as calibration points to capture the driver’s head orientation instead of his/her gaze direction. Based on a self-learning algorithm, the method categorizes the head pose in12different gaze zones and then uses particle filter to estimate driver’s current head pose, gaining the accurate zone according to the confident level. Therefore, the algorithm accomplishes the task of automatic calibration and consequently tracks driver’s gaze direction.The experimental results show that, this automatic method of gaze zone calibration which is without the driver’s engagement can achieve same accuracy as the manual method. The mean error of gaze direction is less than5degree, which meets the needs of safe driving and studying driving behavior.