A REAL-TIME ADAPTIVE LEARNING METHOD FOR DRIVER EYE DETECTION

Introduction

Driver drowsiness is one of the major causes of traffic accidents on road. Monitoring the driver’s vigilance level, and issuing an alert when he/she is not paying enough attention to the road is a promising way to reduce the accidents caused by driver factors. Thus it could become an important part in the development of the advanced safety vehicle. The driver’s facial information, especially the eye status is often believed to give some clues of his/her drowsiness level. The driver eye detection methods based on computer vision use camera to obtain the driver’s facial information, extract the parameters that are believed to be related to the drowsiness level of the driver. Many researchers use Percent of Eyelid Closure (PERCLOS) as an indicator to detect drowsiness. Ishii et al. build a system using driver’s facial expression to reflect the mental status. Ueno et al. described a method using eye open level to detect drowsiness . Other commonly used parameters also include the blinking interval, pupil position, etc .In developing those driver monitoring systems, a reliable real-time driver eye detection method is one of the essential parts. In developing the driver eye detection method, we use a driving simulator and a CCD camera mounted at the back of the steering wheel for capturing the driver’s facial image. The driver was asked to drive normally on a circular course with a constant speed of 80 km/h. The driving simulator and scenario in the test are shown in Figure 1.

To adapt to the variances in eye shape and size of different individuals, the algorithm for eye positioning is composed of learning and non-learning mode. The system starts from the learning mode, in which face detection is firstly performed to narrow the search region. Then contour detection and heuristic rules are used to identify the Region Of Interest (ROI) of the eye. By learning the eye region, a set of images that satisfy the pre-set rules are obtained to form the eye templates. When the number of successful learning exceeds a preset threshold, the learning mode swatches to the nonlearning mode, in which the eye templates obtained from the previous stage are used to get the eye position. If the eye position does not meet certain rules, the method will swatch back to the learning mode to re-learn the eye region.