INTELLIGENT DROWSY EYE DETECTION USING CONTOURLET TRANSFORM AND WEB LOCAL DESCRIPTORS

Abstract

There are various non-driver related reasons for automobiles crashes including street conditions, the climate and the mechanical execution of automobiles.However, a significant number of automobile accidents are caused by driver error. Driver error includes drunkenness, fatigue, and drowsiness. Many factors can affect a driver’s ability to control a motor vehicle, such as natural reflexes, recognition and perception. The diminishing of these factors can eventually reduce a driver’s vigilance level. There have been numerous “raising awareness†campaigns about drowsiness and drunken driving. However, they have been ineffective for the most part. Such accidents not only affect the drowsy drivers, but also any potential victims. Driver Drowsiness Detection, which is used to detect or not a driver is drowsy can use different features i.e. heart rate, eye status etc. This work introduces an alerting process for when the driver fall asleep based on computer vision based mechanism in which eye status has been calculated by extracting frames from the ongoing video. There are various features which has been found in literature i.e. DWT, statistics, LBP etc. but still an improvement was required as the drowsiness detection results was not similar when different classifiers were used. Hence a contourlet transform and web local descriptor based feature set has been proposed which gives high accuracy in drowsiness detection when different classifiers have been tested by this feature set. Also video compression does not affect much when we evaluated this feature set and high accuracy has been achieved by different classifiers