Continuous Vigilance Prediction Using Deep Learning Method

Vigilance is the ability that people can maintain concentration over a period of time.People with high vigilance can respond external stimuli quickly which protect them from dangers.For some occupations such as drivers or doctors,high vigilance is required all the time otherwise tragedies may happen.So it's very important and significant to do vigilance prediction in order to detect danger in advance and prevent accidents happening.Lots of signals can be used as input for vigilance prediction system,take the vigilance prediction of driver for example,EEG,EOG,video and the grip strength on steering wheel all can be used to predict the vigilance level.Different kinds of signals have different modalities and describe the vigilance from different sides.It's difficult to fuse multi modalities when constructing the vigilance prediction system.Recent years,the progress in deep learning area provides us new ideas.In this research,we will explore the way utilizing deep neural network and deep recurrent neural network to fuse EEG and EOG.We will show how to extract high level features from EEG and EOG which can better describe the vigilance level.We also devised the experiments to prove our method can achieve better results than normal vigilance prediction methods.