查詢結果分析
來源資料
頁籤選單縮合
題名 | 智慧型人眼狀態偵測系統=Intelligent Detection System for the State of Human Eyes |
---|---|
作 者 | 林國暐; 陳良驊; | 書刊名 | 龍華科技大學學報 |
卷期 | 30 2010.12[民99.12] |
頁次 | 頁13-26 |
分類號 | 312.13 |
關鍵詞 | 人臉偵測; 眼睛狀態偵測; 膚色偵測; Adaboost; 2DPCA; Human face detection; Eyes state detection; Skin-color detection; |
語文 | 中文(Chinese) |
中文摘要 | 本論文主要是研究並設計一個智慧型人眼狀態偵測系統,用來偵測影像中人眼睛是處於開眼還是閉眼狀態,本系統適用於人臉處於正臉且沒有配戴眼鏡的狀況。論文主要包含三個部份,第一部份是進行人眼狀態資料庫的建立,第二部份是針對人臉進行偵測及膚色驗證,第三部份為人眼部位的切割及狀態辨識與資料分析。人臉偵測的部份,本論文採用Adaboost人臉偵測方法來尋找人臉,但此方法常會有一些誤判的情形,因此我們加入了HSV膚色偵測,來排除誤判的人臉案例,並進一步利用膚色偵測後所 產生的二值影像對眼睛部位的特徵進行搜尋與切割,最後將切割出來的部份進行二維主成份分析 (Two-Dimensional Principal Component Analysis, 2DPCA) 的運算,求出切割影像的影像特徵矩陣 (Image Feature Matrix),再與人眼部位影像資料庫內影像特徵矩陣進行比對,依最小距離法判別該人眼屬於何種狀態。本論文針對三種不同的2DPCA方法進行測試,分別為2DPCA、T-2DPCA、2D2DPCA,實驗測試結果顯示我們可以達到將近九成的成功辨識率。而經比較顯示,2DPCA的辨識正確率優於T-2DPCA和2D2DPCA。運算速度則是2D2DPCA優於2DPCA與T-2DPCA。 |
英文摘要 | An intelligent detection system for the state of human eyes is studied and designed in this paper. This system is used to detect the open or closed state of the human eyes and suitable for the human face images that are under right angle and without wearing spectacles. This study is composed of three parts. The first part is constructing human eye state image database. The second part is human face detection and skin-color verification. The third part is segmentation and state detection of human eyes. For human face detection, we use Adaboost human face detection algorithm to find the face, but there are some results that are false positive for the detected result. We use HSV skin-color detection to exclude the false positive results. Besides, we utilize the binary image of the skin-color image to search the feature of eyes and to segment the eye area. Then we use the segmentation result to perform 2DPCA algorithm and get the image feature matrix. We compare the image feature matrix of the segmented image with the image feature matrices of the eye-state images in the database. According to the minimal distance rule, we can determine the human eyes belonging to which kind of state. Finally, three kinds of 2DPCA algorithms, i.e. 2DPCA, T-2DPCA and 2D2DPCA, are used and compared in this work. From the testing results, the successful rate of recognition is nearly 90%. In addition, the comparison results show that the successful rate of recognition of 2DPCA is better than that of T-2DPCA and 2D2DPCA. On the other hand, 2D2DPCA is better than 2DPCA and T-2DPCA on the time required for detection. |
本系統之摘要資訊系依該期刊論文摘要之資訊為主。