This paper describes a new method of the driver behavior surveillance by using a camera mounted on the test car. Recently

one of the important issues to heavy traffic accidence in the world is that many drivers steer tiredly or are absent-minded when steering car. This situation usually decreases the attention of drivers. Hence

traffic accidence occurs. Thus it is very important to surveille the drivers' behavior

especially the eyes' state recognition to drivers. As a matter of fact

investigation statistics indicates the degree of the drivers eyes' opening is different to some extent on the various occasions of the driver normal driving

the drowsing driving and the fatigued driving. So when the car runs

we obtain two-dimensional image through a on-board camera

from which we can get the eyes' state data. Based on this point

we have developed an effective algorithm to count the frequency of blinkling. This algorithm uses Gabor wavelet to pick up the texture character of the canthus. And the eigenvector that composed of eigenvlaue is the input of the ANN. We also have performed outdoor experiments to make sure that our algorithm works effective.