Driver Fatigue Detection System Based On Face Recognition

Fatigue driving, as a high threat factors threatening the lives of drivers, it has always been the problem that the researchers want to break. Existing driver fatigue detection method to combine software and hardware together. Because of individual differences as well as fatigue and non-fatigue driving difference was not significant, there is no uniform driver fatigue detection rules. This also led driver fatigue detection products are often unsatisfied. The method needs to have physical contact with driver does not give them comfort. A lot of bad real-time detection algorithms, which have led to driver fatigue detection products are not able to get a useful promotion. So far, any mature products for driver fatigue detection is unavailable domestic, research is still at an experimental stage, so the research and develop of the system is valuable.The driver fatigue detection system based on face recognition in the article consists face recognition and driver fatigue detection. The method used for face recognition is stacked aotuencoder from Deep Learning ideaology. It results the individual and eye widen status. The followed is PERCLOS method to identify an individual’s eyes widen status to determine whether the member is driving fatigue. The main work are the following points:1) Mastered the basic theory and the history under the ideaology of Deep Learning and also master the detailed calculation process of stacked autoencoders algorithm. As used herein, the stack autoencoders includes four hidden layers and a classification layer. The nodes in each hidden layer ais 255. 5000 unlabeled human face images as input, training weights four hidden layers, the hidden layer can be used to extract the facial features. Then amount of 500 training set includes labeled human faces, input to the four hidden layer to extract facial features and then input it to the classification layer to train weights for classifiers. Finally, fine-tuning the entire network.2) Algorithm for face recognition detects driver fatigue test successfully and throughout the whole process. Test set of labeled human faces is ordered. It contains 250 human faces. It goes through the successfully trained stacked autoencoders and output the sequence of the same person’s eye open degree. After calculate PERCLOS algorithm, we can draw whether the person is fatigued driving.3) Selection of the layer number and the hidden layer neurons is the result of a lot of experiments, that a time and performance of a balanced choice.4) completed the coding of the entire system. The system can enter the picture for a complete calculation, the completion of face recognition, the human eye state identification, driving status recognition features, as well as follow-up system for parameter update features face recognition algorithm, the driver update function as well as password changes are to be achieved.