Research Of Vigilance Estimation Based On Video Facial Image

Vigilance is usually defned as the ability to maintain attention and alert to ex-ternal stimuli for a long time. Human-computer interaction systems need operatorsto maintain sufcient vigilance. Some special work, such as controllers in air trafccontrol center, pilots and high-speed train or bus driver vigilance is especially impor-tant. Accurate estimate of vigilance and real-time monitoring of it is a very importanttopic in the study of human-computer interaction systems. Methods based on videoinformation due to its portability and low cost has become an important means of vig-ilance estimation. However, because of the ambient light conditions and environmentchanges, the method seems not so outstanding compared with other physiological sig-nals methods. This article gets PERCLOS features by binarization and image gradientmethods and takes advantages of a variety of video image features to propose a novelsystem such as yawn frequency, the eyelid movement speed, body posture ones. Webuildmodelsfordiferentcharacteristicsinsubjectswhichalsoafecttheparametersofmodels to optimize the whole system. This article discusses the gender classifcationproblem based on multi-view facial images. We propose a novel solution using hier-archical classifers structure. The symmetry characteristic of the human face is madeuse of to improve the classifcation accuracy greatly. A variety of diferent classifercombination strategies are experimented. we frmly believe that this method can bewidely applied to prevent accidents like fatigued driving in the future.