利用模糊積分於昏睡偵測

Abstract

當人們在工作中或是在駕駛的環境中，打瞌睡是造成意外事故最常見的因素之一，為了避免類似的意外發生，我們提出了一個非侵入式的昏睡偵測演算法來避免因為侵入式的方法而造成受測者的不舒適感。本文是根據駕駛者眼睛閉合的程度與眨眼頻率兩種偵測資訊來判斷出受測者的昏睡程度。我們首先研究觀察的時間間隔對於利用眼睛閉合程度與眨眼頻率來偵測昏睡狀態的影響，並且找出最佳的觀察時間間隔。為了提高偵測的準確度，我們利用模糊積分的觀念，發展出上述兩種偵測資訊整合的技術，此技術可解決兩種偵測資訊在判斷上發生衝突與模稜兩可的情況。我們也將本文所提出的方法與眼睛閉合程度、眨眼頻率二種方法做比較，根據結果顯示，我們提出的方法的準確率高達95.1%。我們也將所提的方法應用在偵測駕駛者的精神狀態，由結果證明，此實驗是非常成功且有效率的。另一方面，有許多駕駛者有戴墨鏡的習慣，尤其是在夏天，所以我們針對墨鏡的區域來做影像增強以去除墨鏡對眼睛偵測與昏睡偵測的干擾。

Drowsiness is often reported to be one of the most important factors causing danger on various occasions such as work fields and vehicle driving. To avoid this danger, we thesis in this thesis a non-intrusive vision-based drowsiness detection algorithm. Visual techniques are adopted such that we can prevent people from feeling uncomfortable due to intrusive signal acquisition. In this study, we utilize the Long Duration Blink Frequency (LDBF) and the PERcentage of eyelid CLOSure (PERCLOS) as features of drowsiness detection, which are commonly used in visual drowsiness detection system. We first investigated the effect of Observing Time Interval (OTI) on the separability of sample distributions of LDBF and PERCLOS under drowsy and conscious states to select the best OTI. In order to increase the accuracy of drowsiness detection, we use fuzzy integral to combine two different information sources from LDBF and PERCLOS features. The proposed fuzzy integral approach can resolve the conditional unreliability and uncertainty encountered in using LDBF or PERCLOS singly. To show the superiority of our method in drowsiness detection accuracy, we compared our proposed method to LDBF and PERCLOS, respectively. According to the experiment result, the proposed algorithm has the best average detection accuracy of 95.1%. In practice, we also implemented our algorithm to determine people’s vigilance in a driver monitor and warning system. The test in driver drowsiness detection and warning was successful and satisfactory. On the other hand, many drivers have the needs and habits to wear sunglasses, especially in summer. We also develop image enhancement techniques to eliminate the effect caused by sunglasses in eye detection and drowsiness detection.