車輛主動式安全防護之瞌睡偵測與警示

 

Drowsiness Detection and Warning for Vehicle Active Safety

 

蔡依陵、丁肇隆、張瑞益*

國立臺灣大學 工程科學系

羅斯福路四段1

台北市10617大安區

*rayichang@ntu.edu.tw

 

I-Ling Tsai, Chao-Lung Ting, Ray-I Chang*

Dept. of Engineering Science, National Taiwan University

No. 1, Sec. 4, Roosevelt Road

Taipei, 10617, Taiwan

*rayichang@ntu.edu.tw

 

 

摘要

因人員瞌睡所引發之意外事故時有所聞,往往在交通安全上造成很大危害。本論文提出了一個駕駛員瞌睡偵測演算法,只需以一般的攝影與計算元件(:智慧手機)就能達到快速準確的持續追蹤人員眼部資訊,進而判斷駕駛員是否出現瞌睡或其他可能不正常駕駛的情況,發佈警示,以確保行車安全。整個系統主要分為兩個部分,其中偵測子系統負責在系統一開始時進行基本人臉定位,接著進入追蹤子系統來進行瞌睡警示判別。由於系統一開始時已完成基本定位,追蹤子系統可快速微調偵測驗證及定位眼睛區塊,再偵測瞳孔高度並判別眼睛開狀態以偵測瞌睡。系統只有在追蹤失敗一段時間後,才會回到偵測子系統重新進行基本人臉定位,所以正常情況下處理十分快速。而如果人臉偵測數次都失敗,很可能是在不正常駕駛的情況下所引起,系統也會發佈警示。經實驗證明,以目前智慧手機的1GHz CPU360*240攝影解析度,本系統偵測一張影像約需0.33秒,而追蹤一張影像約需0.07秒,其偵測瞌睡的正確率可達99.32%。由於有偵測子系統所提供的正常眼睛開大小做參考,偵測瞌睡的假警報率很低,頗具有實用性。

 

關鍵詞: 瞌睡偵測、車輛主動式安全防護、臉部偵測、眼睛定位。

 

Abstract

Accidents caused by drowsy driving have occurred frequently, which has become the most serious concern of the government and the society. This paper proposes an non-intrusive drowsy driver detection algorithm which is capable to determine drivers' mental state by analyzing the driver's eyes. When system detects the driver's eyes close for a period of time, the alarm will be triggered to ensure safe driving. The system includes five functions: face detection, eye region detection, eye verification, eye location, and drowsiness justification. Two modes, detecting model and tracking model, are included during system operation. The detecting mode is used to locate driver's eyes at first time and verify the eyes are in the detected region. Once successful operation of the detecting mode is achieved, the system is switched to the tracking mode. By the continuous movement of eyes, a small region is searched and the eyes can be located quickly in the tracking mode. Therefore, using 1GHz CPU and 360*240 camera of a modern smart phone, the state of driver’s eyes can be detected in 0.33 second and tracked in 0.07 second. From our experimental results, the proposed system has 99.32% correctness in drowsiness detection is achieved using the proposed method.

 

Keywords: Drowsy Detection, Vehicle Active Safety, Face Detection, Eye Location