Driver Fatigue Monitoring And Warning System Based On Multi-parameter Fusion

With the increase in the number of cars, traffic safety problem has become a difficult problem in the field of international road transportation, with human factor being the main cause of traffic accidents. Driver fatigue as human factor is a significant cause of traffic accidents. We identify the importance of developing a real-time driver fatigue monitoring and warming devices in order to help prevent driving accidents. In recent years, many researchers have been working on the research and development of fatigue monitoring system using different techniques. Fatigue detection technology based on physiological phenomenon applied extensively. EEG is usually regarded as "gold standard" for fatigue detection and people think is the most reliable and meaningful detection method. Research also shows that the brain is a complex nonlinear dynamical system, so the nonlinear dynamics theory provides a new platform for EEG studies. Mixture of experts network has a modular neural network architecture. The idea of mixture of experts network is "to divide for conquer’".That is to say, a complex problem is subdivided into simpler sub problems that are treated independently.Based on the above reasons, this paper extracted three kinds of nonlinear characteristic parameters from EEG as driver fatigue characteristic value. Include Approximate Entropy, Kc complexity and CO complexity and established mixture of experts network classifier model, realizing the forecast of waking state and drowsy. Meanwhile, this paper realized the detection of eye-closed state as an aided detection of drowsy in the wavelet packet domain using power spectrum. The system will make a sound or send a visual indication signal to remind the driver to pay attention to safety driving when detecting the abnormal state. Under the best window after testing, the accuracy of arousal and drowsy state identification is75.25±9.21%; the accuracy of eye-open and eye-closed identification is87.31±3.97%; the average processing time of system monitoring is170.8±12.4ms. In conclusion, the way applied in this paper can quickly and effectively identify eyes-open, eye-closed and drowsy state. The system has the advantages of high degree of modularity, adjustable parameters, extendibility, practicability and provide a good model for the development of real-time and portable fatigue monitoring and warning device.