Many traffic accidents are caused by drowsiness while driving. The purpose of this research is to detect a driver's blinking using facial images obtained by a TV camera and to estimate the driver's arousal level to prevent the occurrence of an accident due to drowsiness. This report presents the method used to obtain the facial images, which are not affected much by the change of the ambient light by day and by night owing to the infrared light sources. This paper also describes the method for extracting the blinks and measuring the blinking durations, which are not influenced very much by individual differences in eye shape. Furthermore, we experiment with the relation between blinking durations and arousal levels as compared with vehicle deviation from the lane and driver's answer to the question about drowsiness. We also examine a method for estimating arousal level which reflects individual characteristics.