The Research of Multi-Modal Method of Drowsy Detection for Driver

• Main Authors: Hsiang-Ming Hsu, 徐祥鳴
• Other Authors: Ning-Han Liu
• Format: Others
• Language: zh-TW
• Published: 2013
• Online Access: http://ndltd.ncl.edu.tw/handle/52696830326057528637

碩士 === 國立屏東科技大學 === 資訊管理系所 === 101 === In recent years, the drivers’ negligence caused traffic accidents in our country from time to time. Most scholars and research institutions think the major reason of traffic accidents is fatigue driving. According to the research which pointed out that the spirit of driving will fall more than two hours, so driving in a long time is one of the main reasons to bring about drowsy driving. Furthermore, the monotonous external environment for the driver in long-distance driving is likely to cause fatigue driving. A lot of researches found that up to more than 40% drivers with driving duty in all fatigue driving accidents. They can not get adequate rest when they feel tired at work, so the probability of traffic accidents is higher than the other peoples’. It also causes a life threat to other human. In this study, we analyzed the driver’s EEG data and adopted Artificial Neural Networks (ANN), Support Vector Machine (SVM) and k Nearest Neighbor (kNN) classification methods to classify brain data into two categories with drowsy and non-drowsy. Furthermore, to increase the accuracy of prediction we integrated the results of the above methods through a fusion function. The parameters in the fusion function of multi-modal method were estimated through a genetic algorithm. To sum up, the classification of brain data was the primary topic in this study. In the experiments of this study, the brain data was collected through a brainwave instrument. The prediction results of the classification methods were compared to each others. We also analyzed the rates of true/negative and false/negative of these methods. According to the experimental results, we found that the classification accuracy rates of three methods are between 70% and 80%. The multi-modal method we proposed can get the highest accuracy rate of classification. Moreover, the experimental results also show that the methods using multiple subjects’ dataset outperform the ones using the personal dataset.