Design And Implementation Of Fatigue Driving Detection System Based On Deep Learning

Car provide a convenient travel way to people.It narrows the distance between people,the distance between the city and the city.At the same time,this modern way of transportation has led to a number of accidents in human society.At night,the driver is more likely to fatigue.This led to 60% of major traffic accidents occurred at night.Therefore,a system judging whether the driver is in a state of fatigue is necessary.We need to use infrared to enhance light at night due to the poor lighting conditions.The SNR of the infrared image is bad.After comparing several main filtering methods,it is found that anisotropic diffusion filter is better than the other two methods in noise reduction and image edge protection.Therefore,anisotropic diffusion filtering is used as a noise reduction method to eliminate the noise inside the image.Infrared light is used to enhance light.It always causes that the eye image intensity is not uniform.That is bad for image segmentation.After comparing the results of Wallis and Mask,the Mask dodging which can better balance the whole illumination intensity is selected.After using OTSU to segment the eye,several eye-surrounding interference factors are usually retained.At this time,we need to identify the eyes according to the eye characteristics.The artificial selection of features has the shortcomings such as randomness,cumbersome debugging process,personal experience.By using the automatic learning feature of deep learning,it avoids the repetitive debugging process of artificial selection,which further improves the recognition rate of the eye.The Caffenet structure is used as the training network.The segmented image is divided into eyes and non-eyes by the method of distinguishing between connected regions and labeled.Use fine-tune to speed up the training process.To do the appropriate conversion for PERCLOS standard,the eye height is as a judge of fatigue standards.The experimental results show that after noise reduction,smoothing,segmentation and recognition,it can identify the eye part image with high accuracy,protect the eye features and improve the accuracy of driver fatigue judgment.