Driver Fatigue Detection Technology Research Based On Computer Vision

With the rapid advances in technology and science, the transportation industry has got a great development, the high rate of road accodents become a challenging issue to human’s safty and property. And an importment factor which causes traffic accidents frequently happen is fatigue driving. Therefore, to decelop a fatigue detection system with high accucacy and intelligence has the extremely meaning.In reference to a lagre number of domestic, this paper chose the machine learning method in computer vision field to achieve the fatigue detection. The driver fatigue detection system contains three parts:improvement of face detection method based on Adaboost algorithm, judgement of fatigue state, the simulation software testing.This paper uses Adaboost algorithm to train face and eye classifier. Considering the disadvantages of traditional algorithm in training time, design an improved FTAdaboost algorithm based on feature tailoring. According to the curve of error rate after initialization, algorithm tailors the features before the first inflection point based on their pre-classification ability, make the training more efficient. In face detection, combine the method of skin-color space, make judgement to the specified face region whether meet the skin-color characteristics, reduce the error rate. In fatigue detection part, improve the way in judging fatigue state by a multi-polarization judgment instead of traditional two-polarization judgment. Make the judgment by calculating the ratio of visible area of iris in defferent eye states and the whole area of iris, detect the fatigue states before the driver sleep as early as possible. Last, give the result by PERCLOS standard. The experiment results show that, the improved FTAdaboost algorithm save lots of training time without the descent of accuracy, and the detection method with skin color space reduce the error rate effectively. Compared with traditional fatigue detection method, the improved fatigue detection method has higher accuary in different fatigue states and also has good real-time ability.