Multi-featured and conditional probability analysis model for driving-disability identification

Abstract

Eye Blink Detection is gazing or the gesture specific analysis utilized in different application area. In this paper, the eye blink frequency as well as pupil movement analysis is observed to identify the chances of fatigue, sleep or drunken person. Driving in such critical situation is a crime. The work is here provided to automate the identification of driving capability of a person by observing the eye characteristic analysis. This designed framework will first identify the eye frames based on the person and eye region localization. This localization will be based on geometric and color model analysis. Now frame similarity observation in eye region will be applied based on statistical measures applied random blocks. As the key frames applied to process the on key eye area. An this stage, a center specific movement observation is applied along with a conditional probabilistic estimation. A rule based conditional estimation will be applied on pupil movement and eye blink. This rule formation is able to identify the degree of person disability to drive the vehicle. The work is applied on CVGLAB videos and the real time captured high resolution videos. The accuracy of eye blink identification is about 88 to 92% for different realtime videos whereas the accuracy on CVGLAB database is about over 93%.