Study On The Recognition Of Eye State Based On Wavelet Transform

With the growing number of motor vehicle holdings, road traffic accidents areincreasing. Fatigue driving is one of the important causes of traffic accidents.Researchers are exploring a monitoring mechanism to detect whether the driver isfatigued. If he is fatigued, it can indicate his dangerous state and help him takecorresponding measures consciously to avoid fatigue driving traffic accidents. Eyeopening and closing state recognition is one of the simple and effective methods fordriver fatigue detection.This paper mainly studies using Gabor wavelet transform to identify eye state. First, It uses skincolor model for face detection to narrow conversion region. Then it extracts the specified scale anddirection relevant features in the frequency domain by Gabor wavelet transform. It puts forward twokinds of effective eye opening and closing state recognition method: eye state recognition based onGabor wavelet transform and integral projection and eye state recognition based on Gabor wavelettransform and pattern recognition. The first method highlights the opening and closingcharacteristics of eyes in the vertical direction in the Gabor image and weakens other organs’ grayfacial features influences by adjusting the parameters of Gabor filter. It puts forward using thepeak-to-average ratio as the criterion for eye state recognition. Because it’s different from each other.The second method makes an integral projection curve of Gabor gray image to find eyes position. Itchooses only a handful of key feature points’ amplitude of Gabor as feature vector. Then it puts thevector into the SVM classifier to create the model and gets the recognition result.The realization of algorithm’s program packages is designed by VC++6.0platform, theOpenCV computer vision library and LIBSVM software. It gives the sample results.Theexperimental results show that the two methods can identify eye opening and closing stateeffectively. The first method does not need the eye accurate position. Its realization is simple. Thefeature vector dimension of the second method is low and do not need to reduce.This method has ahigh recognition rate. The two methods can both be applied to the fatigue driving monitoringmechanism. They can prevent fatigue driving traffic accidents.