【摘要】：眼睛是感知世界的重要器官,視線方向可以反映人們感興趣的點。視線跟蹤技術(shù)可以檢測出人類眼睛的注視方向,得出興趣點。隨著電子技術(shù)的不斷發(fā)展,關(guān)于視線研究技術(shù)的研究越來越多,一些視線跟蹤系統(tǒng)已經(jīng)應(yīng)用于人機(jī)交互領(lǐng)域。隨著研究的不斷深入,視線跟蹤技術(shù)在網(wǎng)絡(luò)的可用性、廣告、包裝設(shè)計和汽車工程等領(lǐng)域也會有很大的發(fā)展空間。但是現(xiàn)有的視線跟蹤技術(shù)有限,視線跟蹤系統(tǒng)存在跟蹤精度較低、限制頭部運動、干擾性大的缺點。針對上述問題,論文對序列內(nèi)的視線跟蹤技術(shù)進(jìn)行了研究,減小了對使用者頭部的限制,提高了系統(tǒng)的準(zhǔn)確度和穩(wěn)定性。 論文的主要工作成果如下： (1)把頭部姿態(tài)估計方法與二維視線跟蹤方法結(jié)合,利用頭部姿態(tài)參數(shù)校正用于視線估計的面部特征點間的位置距離,減小因頭部發(fā)生轉(zhuǎn)動給視線跟蹤算法帶來的誤差。該方法既不需要輔助設(shè)備固定頭部,又提高了視線跟蹤的準(zhǔn)確度。 (2)對用于視線跟蹤的頭部姿態(tài)估計算法進(jìn)行了研究,提出了一種三維頭部姿態(tài)估計方法。在該算法中,把頭部看作圓柱體,頭部的轉(zhuǎn)動可以看作是圓柱體的旋轉(zhuǎn)。通過不斷變換姿態(tài)參數(shù),使得當(dāng)前的面部紋理與參照紋理相符合,此時的參數(shù)即為當(dāng)前圖像的頭部姿態(tài)參數(shù)。利用視頻圖像幀間頭部圖像變化較小的特點,利用前一幀圖像的頭部紋理估計下一幀圖像頭部的姿態(tài),減少了計算量,提高了計算精度。 (3)在邊緣定位方法中對亞像素技術(shù)進(jìn)行了深入的研究。在視線跟蹤過程中,利用亞像素技術(shù)定位面部特征點(虹膜中心和外眼角)的亞像素位置；根據(jù)檢測得到的虹膜亞像素級邊緣點,利用橢圓擬合方法精確定位虹膜亞像素中心點。該方法減小因二維面部圖像特征點定位不夠精確給視線方向估計帶來的誤差,提高了視線跟蹤的準(zhǔn)確度。 (4)采用簡單的頭部幾何模型,提前采集眼睛位于屏幕固定位置時的面部特征點,根據(jù)這些特征點組成的向量與視線方向之間的對應(yīng)關(guān)系,估計當(dāng)前頭部圖像中校正后的特征點向量的視線方向。該方法計算簡單且能夠準(zhǔn)確估計視線方向,使序列內(nèi)視線跟蹤系統(tǒng)能夠滿足實時性要求。實現(xiàn)了視頻序列內(nèi)視線跟蹤系統(tǒng),驗證了系統(tǒng)的準(zhǔn)確度和穩(wěn)定性。
[Abstract]:The eye is an important organ for perceiving the world, and the direction of sight can reflect the point of interest. Eye tracking technology can detect the gaze of the human eye and get the point of interest. With the development of electronic technology, more and more researches have been made on the line of sight technology, and some line of sight tracking systems have been applied in the field of human-computer interaction. With the development of the research, the technology of line of sight tracking will have great development space in the field of network usability, advertising, packaging design and automobile engineering. However, the existing line of sight tracking technology is limited, and the tracking system has the disadvantages of low tracking accuracy, limited head movement and large interference. Aiming at the above problems, this paper studies the line-of-sight tracking technology in the sequence, reduces the limitation on the user's head, and improves the accuracy and stability of the system. The main achievements of this paper are as follows: (1) combining the head attitude estimation method with the two-dimensional line of sight tracking method, the position distance between the facial feature points used for the line of sight estimation is corrected by using the head attitude parameters. The error caused by the rotation of the head to the line of sight tracking algorithm is reduced. This method not only needs no auxiliary equipment to fix the head, but also improves the accuracy of line of sight tracking. (2) the head attitude estimation algorithm for line of sight tracking is studied, and a three-dimensional head attitude estimation method is proposed. In this algorithm, the head is regarded as a cylinder, and the rotation of the head can be regarded as the rotation of the cylinder. By constantly changing the attitude parameters, the current facial texture is consistent with the reference texture, which is the head pose parameter of the current image. Based on the small change of the head image between the video images, the head texture of the previous frame image is used to estimate the pose of the head image of the next frame, which reduces the computation cost. The calculation accuracy is improved. (3) the sub-pixel technique is studied in the edge location method. In the course of line of sight tracking, the sub-pixel position of facial feature points (iris center and outer eye corner) is located by sub-pixel technique. The ellipse fitting method is used to locate the center of the iris subpixel accurately. This method reduces the error caused by the location of feature points in two-dimensional facial images and improves the accuracy of line of sight tracking. (4) A simple head geometry model is used. According to the relationship between the vector of these feature points and the direction of line of sight, the direction of line of sight of the corrected feature point vector in the current head image is estimated according to the relationship between the vector and the direction of line of sight. The method is simple in calculation and can accurately estimate the direction of line of sight, so that the system can meet the real-time requirements. The video sequence line of sight tracking system is implemented, and the accuracy and stability of the system are verified.
【學(xué)位授予單位】：武漢理工大學(xué)
【學(xué)位級別】：碩士
【學(xué)位授予年份】：2011
【分類號】：TP391.41

【參考文獻(xiàn)】

相關(guān)期刊論文 前4條

1 于瓊;樊銀濤;;多角度視線上的人眼定位算法[J];電腦知識與技術(shù);2011年04期

2 張文聰;李斌;鄧宏平;譚文明;莊鎮(zhèn)泉;;視線跟蹤過程中變形瞳孔的定位[J];電子與信息學(xué)報;2010年02期

3 鐘志光;易建強(qiáng);趙冬斌;洪義平;李新征;;一種魯棒的只需兩幀圖像的姿態(tài)估計方法[J];模式識別與人工智能;2005年04期

4 徐蓉;姜峰;姚鴻勛;;流形學(xué)習(xí)概述[J];智能系統(tǒng)學(xué)報;2006年01期

相關(guān)博士學(xué)位論文 前1條

1 山世光;人臉識別中若干關(guān)鍵問題的研究[D];中國科學(xué)院研究生院（計算技術(shù)研究所）;2004年

相關(guān)碩士學(xué)位論文 前2條

1 李華明;基于中國人面貌形態(tài)學(xué)特征的人臉姿態(tài)估計方法研究[D];西北大學(xué);2005年

2 孫昊翔;關(guān)于人臉姿態(tài)估計問題的研究[D];吉林大學(xué);2007年

，

本文編號：2125093