Facial Features and Eyes Closure Dependent Intelligent Method for Driver Drowsiness Identification: Multiple Kernel Learning Based Approach

Abstract

Fatigue, lack of concentration or drowsiness of drivers is one of the prominent reasons for roadside accidents world-wide. To address this problem, we have proposed a method to detect drowsiness of the driver based on complete eyes closure including the head position, & facial fatigue as conveyed by yawning and a drowsy face. Eye closure is calculated using landmarks on eyes. Detection of facial expressions of fatigue is divided into three phases. In the first phase, a Curvelet Transform is implemented to transform the input face image into four sub-band images which retain significant facial expression and eye information. The initial image is also sampled to acquire images of different sizes. Based on entropy assessment, each image is further divided into several blocks that are either categorized as informative or non-informative. In the second phase, using Discrete Cosine Transform the characteristics of high variance are selected in zigzag form. In the final phase, the Multiple Kernel Learning classifier is trained and tested to properly classify fatigue expressions of the driver. We combined eye aspect ratio (eye closure) with fatigue facial expressions (yawning and drowsiness) for drowsiness detection of the driver. It is observed that the proposed method has better classification accuracy and low false positive rate in real time detection than the existing methods.

 Keywords: Entropy analysis, facial recognition, low-resolution, drowsiness detection, curvelet transform, eye aspect ratio