Research On Detection Method Of Driver Fatigue State Via KFEP Algorithm

In China, with the continuous growth of economy and the development of road traffic, the total numbers of vehicles and drivers increase accordingly, leading to the gradual increase of potential traffic safety problems. Among the various reasons of traffic accidents, the driver fatigue is an important one. Real-time detecting drivers and alarming the fatigue ones can promote them to stop fatigue driving as soon as possible, which is extremely significant to reduce the accidents due to driver fatigue and to protect the personal and property safety of the drivers themselves and others.Among the various fatigue detection methods, PERCLOS algorithm judges the fatigue state of the drivers basing on the ratio of their eyes closing time, and has a high degree of correlation with the driver fatigue state. However, it has certain limitations and shortcomings. This thesis researches PERCLOS algorithm and improves it into a new fatigue detection algorithm, and its main works and achievements are as follows:(1) The original PERCLOS algorithm has a difficulty in selecting the threshold value for judging fatigue according to the blink frequency, since different drivers have different frequencies. Pointing to this problem, this thesis adopts the filtering method to solve it. After a comprehensive analysis and comparison of a variety of filtering algorithms, Kalman filter algorithm is considered as the most appropriate and thus is adopted to filter the effect of the blink of the drivers when they are not fatigue. As the result of the research, the proposed Kalman Filter Enhanced PERCLOS algorithm realizes an improvement of the fatigue state detection methods.(2) In this thesis, a hierarchical approach is adopted to realize the extract of fatigue characteristics values. Firstly, the Haar-like features based Adaboost algorithm is adopted to detect and locate the human face. Secondly, the human eye is roughly located depending on the fixed position of human eye. Then, comparing with and analyzing multiple binary image segmentation methods as well as joining with the characters of human eye, the image binarization method of Bisecting Local Means Clustering is proposed to process the binary segmentation of the human eye and to locate the human eye exactly. Finally, basing on the obtained human eye binary image, the maximum height difference between the upper and lower eyelids is calculated in order to obtain the desired fatigue characteristic values.(3) Through simulation experiments, the research content of this thesis has been realized and tested. Firstly, the optimal parameters are selected for the fatigue feature extraction method, and then simulation experiments are processed basing on these parameters. The experimental results prove that the fatigue characteristic values extraction method researched in this paperthesis is effective. Then, basing on this fatigue feature extraction method, the optimal Kalman filter parameters are selected, and the effects toward PERCLOS algorithm before filter and that after filter are compared with and analyzed. The experimental results prove that the Kalman Filter Enhanced PERCLOS algorithm overcomes the defect of the original PERCLOS algorithm, so it can realize the detection of fatigue state much better.