可應用於一般課堂環境中之人眼開闔狀狀態研究
No Thumbnail Available
Date
2013
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
眼睛開闔辨識是電腦視覺的一個重要技術，能夠在生活中發展成多種應用，大部分的眼睛狀態偵測，環境皆屬於背景較為單純、近距離以及頭部晃動不大的情形，像是汽車駕駛疲勞偵測系統，然而本研究希望能將眼睛開闔辨識應用於一般課堂環境中，因此需要解決在有光線干擾及遠距離低解析度下的環境中，仍能快速且有效辨識眼睛的開闔狀態。
本研究之方法共分成三個部分，分別是人臉偵測、眼睛區域決策，最後則是眼睛狀態辨識。首先對影像做人臉偵測，接著將做完前處理的臉部影像利用局部取像的方法得到眼睛的大致位置，再利用水平投影及垂直投影找出眼睛精確的範圍及位置，最後本研究利用開闔眼睛影像輪廓複雜度之差異設計一套新的特徵擷取方式，並搭配已事前訓練過的SVM模型來判斷眼睛的開闔狀態。
無論是近距離或是遠距離實驗，由實驗結果可證明出在相同的辨識率下，本研究所設計之特徵擷取方式比複雜度函數的方法能判斷出的開閉眼資料比例多，因此整體的執行時間可以降低，也證明了本篇方法的可用性，除了開閉眼整體辨識率皆可達到84.9%以上，且隨著門檻值的調整，執行時間也可比單純用SVM快了1.5至3倍，時間上的減少能帶給本系統很大的效益。
Eye state recognition is an important technology in the computer vision. It can be developed to variety applications. Most eye state recognition is pure background, short distance, and the head does not shack. Due to the application in the general classroom that is light interference and long distance, the purpose of our research is to recognize the eye state quickly and effectively. Our method is divided into three parts, face detection, eye region decision, and eye state recognition. First is to find out the face image and do the pre-processing, then make use of the area of interest (AOI) to get the roughly eye position, the last step is utilizing the horizontal projection and vertical projection to get the precise eye position. Eye state recognition is using our proposed method that is a new way to extract feature from binary image and work with SVM model to determine the eye state. The experiment shows that our proposed method that is a new way to extract feature from binary image is better than complexity function method. And our method is not only performs well in the recognition rate but also in the execution time that is 1.5~3 times faster than SVM method.
Eye state recognition is an important technology in the computer vision. It can be developed to variety applications. Most eye state recognition is pure background, short distance, and the head does not shack. Due to the application in the general classroom that is light interference and long distance, the purpose of our research is to recognize the eye state quickly and effectively. Our method is divided into three parts, face detection, eye region decision, and eye state recognition. First is to find out the face image and do the pre-processing, then make use of the area of interest (AOI) to get the roughly eye position, the last step is utilizing the horizontal projection and vertical projection to get the precise eye position. Eye state recognition is using our proposed method that is a new way to extract feature from binary image and work with SVM model to determine the eye state. The experiment shows that our proposed method that is a new way to extract feature from binary image is better than complexity function method. And our method is not only performs well in the recognition rate but also in the execution time that is 1.5~3 times faster than SVM method.
Description
Keywords
人臉偵測, 眼睛偵測, 眼睛開闔辨識, 熵, 雜度度函數, face detection, eye detection, eye state recognition, Entropy, Complexity Function