本文選題：增強(qiáng)現(xiàn)實(shí)　切入點(diǎn)：虛擬眼鏡試戴　出處：《大連海事大學(xué)》2017年碩士論文　論文類型：學(xué)位論文

【摘要】：近些年電商迅猛發(fā)展,虛擬試戴技術(shù)應(yīng)運(yùn)而生。本文通過將虛擬眼鏡結(jié)合到視頻流中,利用增強(qiáng)現(xiàn)實(shí)技術(shù),達(dá)到虛擬眼鏡試戴的效果。使用戶可以實(shí)時(shí)進(jìn)行人機(jī)交互,用戶的購鏡體驗(yàn)得到提升。本文主要的研究工作如下:在人臉檢測(cè)階段,本文采用基于Hog特征的AdaBoost級(jí)聯(lián)分類器與SVM分類器組合的人臉檢測(cè)方法,該方法在降低漏檢率的同時(shí)也降低了誤檢率,使人臉檢測(cè)的準(zhǔn)確率得到提高。在人臉對(duì)齊階段,采用魯棒性較好的SIFT算法做局部特征提取,結(jié)合SDM算法對(duì)特征點(diǎn)對(duì)齊迭代求精,訓(xùn)練出更加精確的特征點(diǎn)對(duì)齊模型,人臉對(duì)齊的精度相對(duì)傳統(tǒng)ASM方法更高。本文提出了基于姿態(tài)估計(jì)的虛擬眼鏡處理。并從兩個(gè)階段進(jìn)行介紹,首先對(duì)頭部進(jìn)行姿態(tài)估計(jì)。由于本文采用基于幾何投影的姿態(tài)估計(jì)方法,在進(jìn)行姿態(tài)估計(jì)的過程中,使用L-M優(yōu)化算法解非線性最小二乘法,擬合性更好,姿態(tài)估計(jì)更加準(zhǔn)確。在虛擬眼鏡處理階段,本文提出利用姿態(tài)估計(jì)得出的旋轉(zhuǎn)角度,對(duì)眼鏡圖像進(jìn)行適當(dāng)?shù)耐敢曌儞Q,使其在呈現(xiàn)出三維效果的同時(shí),避免了夸張變形的狀況。將眼鏡腿與眼鏡組合,在頭部旋轉(zhuǎn)時(shí),根據(jù)旋轉(zhuǎn)角度計(jì)算出眼鏡腿的顯示長度,給用戶以三維立體的眼鏡試戴效果。最后本文將虛擬眼鏡試戴技術(shù)應(yīng)用到眼鏡試戴系統(tǒng)中,設(shè)計(jì)完成一個(gè)簡單的虛擬眼鏡試戴系統(tǒng)。通過實(shí)驗(yàn)發(fā)現(xiàn),本文采用的虛擬眼鏡試戴技術(shù)可達(dá)到用戶的要求,并能展現(xiàn)出令人滿意的試戴效果。
[Abstract]:In recent years, the rapid development of e-commerce, virtual wear technology came into being. This paper combines virtual glasses into video stream, using augmented reality technology, to achieve the effect of virtual glasses, so that users can real-time human-computer interaction. The main research work of this paper is as follows: in the phase of face detection, we adopt the combination of AdaBoost cascade classifier based on Hog features and SVM classifier. This method not only reduces the false detection rate, but also reduces the false detection rate, which improves the accuracy of human face detection. In the phase of face alignment, the robust SIFT algorithm is used for local feature extraction, and the SDM algorithm is used to align the feature points for iterative refinement. A more accurate feature point alignment model is trained, and the accuracy of face alignment is higher than that of traditional ASM method. In this paper, a virtual spectacle processing based on attitude estimation is proposed. Firstly, the attitude estimation of head is carried out. Because of the geometric projection based attitude estimation method, the L-M optimization algorithm is used to solve the nonlinear least square method in the process of attitude estimation, and the fitting property is better. Attitude estimation is more accurate. In the phase of virtual glasses processing, this paper proposes to use the rotation angle obtained by attitude estimation to transform the spectacle image into perspective properly, so that it can present a three-dimensional effect at the same time. Avoid exaggerated deformation. Combine the glasses leg with the glasses. When the head rotates, calculate the display length of the glasses leg according to the rotation angle. Finally, this paper applies virtual glasses to the system, and designs a simple virtual glasses trial wear system. The virtual spectacles in this paper can meet the needs of users and show satisfactory results.
【學(xué)位授予單位】：大連海事大學(xué)
【學(xué)位級(jí)別】：碩士
【學(xué)位授予年份】：2017
【分類號(hào)】：TP391.41;TP391.9

【參考文獻(xiàn)】

相關(guān)期刊論文 前8條

1 黃華盛;楊阿慶;;基于PCA算法的人臉識(shí)別[J];電子科技;2015年08期

2 盧洋;王世剛;趙文婷;武偉;;基于人臉姿態(tài)估計(jì)的虛擬眼鏡試戴技術(shù)[J];中國光學(xué);2015年04期

3 董力賡;陶霖密;徐光yP;;基于二階梯度朝向直方圖特征的頭部姿態(tài)估計(jì)[J];清華大學(xué)學(xué)報(bào)(自然科學(xué)版);2011年01期

4 張鴻燕;耿征;;Levenberg-Marquardt算法的一種新解釋[J];計(jì)算機(jī)工程與應(yīng)用;2009年19期

5 張勇;金學(xué)波;;基于先驗(yàn)信息和射影幾何變換的車牌圖像矯正[J];計(jì)算機(jī)應(yīng)用研究;2008年07期

6 張振國;童天添;;眼鏡行業(yè)計(jì)算機(jī)選鏡及虛擬試戴技術(shù)研究[J];微計(jì)算機(jī)信息;2007年30期

7 馬曉麗,張桂華;透視圖反求作圖方法的計(jì)算機(jī)仿真算法研究[J];計(jì)算機(jī)仿真;2003年05期

8 金升平,曾梅蘭;仿射變換中點(diǎn)匹配的篩選基的對(duì)準(zhǔn)算法[J];計(jì)算機(jī)工程與應(yīng)用;2003年10期

相關(guān)碩士學(xué)位論文 前2條

1 崔少奇;人臉姿態(tài)估計(jì)算法的研究[D];西北大學(xué);2015年

2 張景森;基于視覺的虛擬眼鏡試戴系統(tǒng)開發(fā)[D];杭州電子科技大學(xué);2015年

，

本文編號(hào)：1642896