The Design And Implementation Of Fatigue Driving Detection System

With the continuous development of society and economy, transportation industryis developing rapidly. The number of cars in China is also growing. Mean while,China’s vicious traffic accident has increased dramatically and has a rising trend.Among them, accidents caused by driver fatigue accounts for20%of the total, whichamounted to more than40%of serious accidents. It is clear that, the harm of fatiguedriving is serious, the research of fatigue detection technology is very necessary, andhas profound significance.At present, the detection of driver fatigue based on computer vision is a hotspot ofresearch. It has been proved that judging driver fatigue by eyes is an effective method.Therefore, basing on the achievements of the predecessors, this paper designed a fatiguedriving detection system. The system has been proved to be high real-time and accuracyrate through many experiments and it’s an applicable navigation system. The maincontents of this paper are as follows:1. This paper studied the face detection and face location technology. Consideringthe real driving environment, this paper uses Adaboost face detection algorithm,whichbased on Haar features, to locate face in video frame. Considering its fast detectionspeed, strong robustness, good real-time performance, I use this method to detect face inthe face detection module.2. This paper studied the technology of eye location and tracking. A new methodhas been designed for eye detection: using classifier based on Haar and Adaboost todetect eyes after a coarse positioning step of the facial region. It not only ensured thereal-time requirements of the system but also improved the accuracy of eye location.Besides, this paper attempts to introduce TLD algorithm for eye tracking module, butthe effect of tracking and real-time performance is not very ideal.3. This paper studied the technology of eye state recognition. This paper presents abinarization method for eye image by multi-threshold. It can represent in3steps:1)segment the eye image using multi-threshold,2)select a image with the optimal contour,3)determine the eye state. Experiments show that this method has a fast speed and has almost no effect by light. It has very good practical value using in the vehicle.