Automatic Calibration of the Driver's Gaze Area Based on a Semi-Supervised Learning Algorithm

Driver's attention can be determined by monitoring the gaze area, which could be used to provide effective warnings. In practice, the height differences in drivers can affect the relative position between the camera and driver, which may increase the misrecognition of the gaze area.Aiming at resolving this problem, an automatic calibration method based on semi-supervised learning was proposed in the present study. To this end, the yaw and pitch of gaze were monitored based on the supervised learning model L2CS(L210ss cross-entropyloss softmax layer) network, and the line-of-sight reference frame was constructed by manual annotation. Then the highest density line-of-sight point area was calculated using the sliding window to complete the clustering. The performed analyses demonstrated that the performance index of gaze area monitoring significantly improves after calibration. It was found that the algorithm has promising non-perception and online repair characteristics,and the accuracy of gaze area classification is increased by 16% after using this calibration method.