駕駛者眼睛的眨眼偵測

Abstract

交通事故死亡率總是佔國內前十大死因之一，其中肇事原因之一是因為駕駛者精神不集中而發生交通意外。故許多學者積極地在車內裝設偵測眨眼頻率的設備，以便預測駕駛者的精神狀態，若是想打瞌睡的時候，趕緊發出警告喚醒駕駛者。 本論文的目的在於使用一般攝影設備拍攝駕駛中的駕駛者，克服光線變化的問題即時從影像中偵測駕駛者眨眼狀況，以應用於偵測駕駛者的精神狀態。在偵測駕駛者眨眼狀況的過程中，主要可分成四步驟：臉部偵測、眼睛偵測、眼睛追蹤與眨眼偵測。首先從輸入的影像中，結合不同來源的色度偵測膚色的區域，快速地擷取駕駛者的臉部，並且使用邊線的資訊修復這個臉部，之後使用橢圓的資訊找出臉部真正的位置。在偵測到駕駛者臉部後，依據人臉比率找出眼睛的範圍，並在這範圍內根據膚色的資訊、眼睛的特徵以及連續數張影像的輔助，準確偵測出駕駛者的眼睛部分。接著使用眼睛的形狀與位置，在連續影像中追蹤眼睛的位置，並且同時利用顏色的資訊判斷駕駛者的眨眼狀況。

Since car accidents may be caused by a driver's drowsiness, some assistance systems have been designed to warn the driver before he falls asleep. Such a system could monitor the driver’s eyes to detect blinking. If the blinking becomes to frequent or prolonged, then it may indicate drowsiness. We propose a vision-based eye blinking detection system in this paper. Four steps, extracting a human face, detecting eye location, tracking the eyes and detecting blinking are developed for our system. Using a video camera mounted in the image of the driver’s face is first extracted based on skin color. Second, the location of the eyes of the driver is detected according to eye features. Third, we track the eyes’ location in next frame based on the shape and the location of human eyes relative to the face. Finally, we detect eye blinking of the driver. Our system has been tested during actual driving, and shows that it can be adapted to changing illumination while the car is in motion.