| Fatigue driving is one of the most important contribution to traffic accident, which makes it pregnant to take research on real-time drowsy detection system. A real-time fatigue driving detection system could prevent drivers who are in drowsy state, and then decrease traffic crashes.In the fatigue driving detection system, the algorithm for driving fatigue is the essential component. Due to reason that the physiology-based detection is intrusive and the performance-based detection is easily distracted, this thesis aims to take research based on driver behavior and vehicle state. This study examines the change of driver behavior and vehicle state under fatigue and designs the identification algorithm for driver fatigue.Firstly, in order to get the parameters needed, a driving experiment based on driving simulator is carried out. These parameters include vehicle’s yaw angle, lane position, speed variation and so on. In this experiment, the data of twelve volunteers are used to train the algorithm, while the data of another two volunteers are used to test the algorithm. Meanwhile, according to SSS self-access learning, the volunteers’ statues are classified into two levels:normal and fatigue. Then the database of fatigue sample is established.Secondly, according to a thorough analysis on the data, the characters which can distinguish driver’s statue are found. Based on existing researches,117 features are extracted. The differences of each feature under different statue are tested by the analysis of variance (ANOVA). Then according to the analysis, seven features are further extracted, which constitutes the input set of algorithms.Finally, on the basis of the feature vectors, the driver fatigue features in the input set were abstracted by principal component analysis (PCA). Then Fuzzy cluster method (FCM) and Artificial neural network (ANN) are used to identify the sample’s statue. The identification effect of different methods is compared with each other. Then, the data of another two drivers are used to test the effectiveness of designed algorithm. |
