Study Of Fatigue State Recognition Based On Multiple Information Fusion

As one of the most important predisposing factors, fatigue driving is a serious nuisance to the driver’s life and property safety. The research about identification, detection and warning of fatigue driving has become a core issue in traffic safety field.At the present stage, the research directions of fatigue status identification can be summarized into three categories:firstly, the identification method based on traditional physiological information has the highest recognition result, but it also affects drivers’routine work by bodies’contact measurement. Secondly, the identification method based on traditional facial image features doesn’t affect people’s normal action, but it can be easily influence by external environment and leads to an abnormal recognition accuracy rate. Finally, the identification method based on vehicles’operating status has the advantages of simple structure, low cost, strong real-time, but it also has disadvantages of poor anti-jamming, low reliability at the same time.Aiming at the problems such as lacking of stability due to single source, and easily influenced by external environment, this article will propose a new method of fatigue state monitoring algorithm based on multi-information fusion, combined with machine learning, image processing and physiological signal research. The model fuses image and pulse information source, extracts the characteristics of the eye state in image and pulse signal in time and frequency domain, and establishes a fusion feature vector space which combined two kinds of characteristics. And finally, building the fatigue state recognition decision by using Principal Component Analysis and Support Vector Machine to achieve the detection and identification of fatigue state. Main works are as follows:(1)Designing a simulated driving fatigue state excitation experiment, and collecting the data as well, the research has considered various problems in the course of the experiment, it built a natural experimental environment, and it has selected accurate, reliable data acquisition equipment and representative candidates. Meanwhile, the research carry out the state excitation, simulation driving and data collection rigorously and objectively.(2)Using the calculation of eyes’state based on Skin color segmentation and Gray scale integral projection method. The algorithm firstly recognize the face in image based on Skin color segmentation method, then positioning eyes’isolation by Gray scale integral projection, finally calculating the characteristics that reflect status of eyes such as PERCLOS, opening degree and blink frequency.(3)Calculating the time and frequency domain characteristics of pulse signal based on Wavelet transform method. De-noising, filtering and preprocessing the pulse signal based on db6 wavelet’s deconstruction and reconstruction of the implementation, and then seek out the wave characteristics of the pulse signal after filtering through Adaptive threshold method. Finally, extracting pulse signal in time and frequency domain such as AVNN, SDNN, HF, LF and multiple characteristic parameters after recognizing the peak point.(4)Establishing decision-making machine of fatigue state recognition based on information fusion, Principal Component Analysis and Support Vector Machine. As to the two kinds of characteristics above-mentioned, the article builds fusion feature vector space, transforms the 22 dimensional signal feature vector to an 8 dimensional principal component eigenvectors through PCA, establishing the decision-making machine of fatigue state recognition by SVM training. And at last, the article verifying, evaluating and analyzing the identification effect of the whole algorithm model from multiple angles, based on the previous feature vector data obtained from the calculation all above.