| 摘要: | 在我國，疲勞駕駛、分心而未注意車前狀況等人為疏失情況是交通事故發生的重要原因，因此本研究目的在開發一套即時且非接觸式的高效能駕駛狀態監控系統，透過影像處理方式即時偵測駕駛人的面朝方向與眼睛開闔狀態，進而精確判斷出駕駛人的疲勞與分心狀態。本系統將攝影機輸入的畫面，透過移動偵測、邊緣偵測、區域亮度偵測、樣版比對、連通域標記演算法、區域特徵分析等影像處理方法，在動態畫面中即時追蹤頭部和眼睛的所在位置，再利用頭部中心與兩眼相對位置來判斷駕駛者面朝方向，並在眼睛區域內分析其特徵，判斷眼睛開闔狀態。最後，本系統以面朝方向角度資訊來判斷駕駛者的分心狀態，並以單位時間內的閉眼時間比例(PERCLOS)來判斷駕駛者的疲勞狀態。 本研究除了完成系統驗證外，更進行多種臉部外型(眼鏡、膚色、髮型、頭部大小)及實車環境(背景複雜度、環境亮暗、對向車眩光)等因素的測試，發現本系統在多數臉部外型及環境情況下，判斷面朝方向的角度誤差都在4°之內，而閉眼辨識率也都在99%以上。最差的情況下，面朝方向角度誤差也在8.5°以內，閉眼辨識率則在78.85%以上，證實本系統於不同臉部外型和背景環境下都能有良好的辨識能力。未來本系統將結合車內影音或座椅及方向盤振動等警示裝置，有效的提醒駕駛人，以避免事故的發生。 Some human errors such as driver fatigue and distraction are the main reason for traffic accidents in Taiwan. In this thesis, we develop a real-time, non-contact and high-Performance driver status detection system. Through image analysis, the system can estimate the state of the driver, such as distraction and fatigue. The system tracks the driver's head and eyes by motion detection, edge detection, brightness detection, template matching, connected component labeling algorithm and feature analyzing. Then determine the direction of the driver's face by Position of head and eyes. Finally, the system calculates the direction of the driver's face and PERCLOS to determine the status of the driver. In this study, we test the system in many factors including eyeglasses, skin color, hair, head size and complexity of the background, brightness and backlight on a real car. We found that in most cases, the errors of facing direction are within 4°, and the detection rate of eyes closed in more than 99%. In the worst case, the errors of facing direction are within 8.5°, and the detection rate of eyes closed in more than 78.85%. It confirmed that the system has good recognition ability. In the future, we will combine audio or vibration alert device to avoid traffic accidents. |
