外文翻譯--基于pac的實時人臉檢測和跟蹤方法

1、<p><b>　　譯文一</b></p><p>　　基于PAC的實時人臉檢測和跟蹤方法</p><p><b>　　摘要：</b></p><p>　　這篇文章提出了復(fù)雜背景條件下，實現(xiàn)實時人臉檢測和跟蹤的一種方法。這種方法是以主要成分分析技術(shù)為基礎(chǔ)的。為了實現(xiàn)人臉的檢測，首先，我們要用一個膚色模型和一些動作

2、信息(如：姿勢、手勢、眼色)。然后，使用PAC技術(shù)檢測這些被檢驗的區(qū)域，從而判定人臉真正的位置。而人臉跟蹤基于歐幾里德（Euclidian）距離的，其中歐幾里德距離在位于以前被跟蹤的人臉和最近被檢測的人臉之間的特征空間中。用于人臉跟蹤的攝像控制器以這樣的方法工作：利用平衡/（pan/tilt）平臺，把被檢測的人臉區(qū)域控制在屏幕的中央。這個方法還可以擴(kuò)展到其他的系統(tǒng)中去，例如電信會議、入侵者檢查系統(tǒng)等等。</p><p

3、><b>　　1.引言</b></p><p>　　視頻信號處理有許多應(yīng)用，例如鑒于通訊可視化的電信會議，為殘疾人服務(wù)的唇讀系統(tǒng)。在上面提到的許多系統(tǒng)中，人臉的檢測喝跟蹤視必不可缺的組成部分。在本文中，涉及到一些實時的人臉區(qū)域跟蹤[1-3]。一般來說，根據(jù)跟蹤角度的不同，可以把跟蹤方法分為兩類。有一部分人把人臉跟蹤分為基于識別的跟蹤喝基于動作的跟蹤，而其他一部分人則把人臉跟蹤分為基于邊

4、緣的跟蹤和基于區(qū)域的跟蹤[4]。</p><p>　　基于識別的跟蹤是真正地以對象識別技術(shù)為基礎(chǔ)的，而跟蹤系統(tǒng)的性能是受到識別方法的效率的限制?；趧幼鞯母櫴且蕾囉趧幼鳈z測技術(shù)，且該技術(shù)可以被分成視頻流（optical flow）的（檢測）方法和動作—能量（motion－energy）的（檢測）方法。</p><p>　　基于邊緣的（跟蹤）方法用于跟蹤一幅圖像序列的邊緣，而這些邊緣通常是

5、主要對象的邊界線。然而，因為被跟蹤的對象必須在色彩和光照條件下顯示出明顯的邊緣變化，所以這些方法會遭遇到彩色和光照的變化。此外，當(dāng)一幅圖像的背景有很明顯的邊緣時，（跟蹤方法）很難提供可靠的（跟蹤）結(jié)果。當(dāng)前很多的文獻(xiàn)都涉及到的這類方法時源于Kass et al.在蛇形匯率波動[5]的成就。因為視頻情景是從包含了多種多樣噪音的實時攝像機(jī)中獲得的，因此許多系統(tǒng)很難得到可靠的人臉跟蹤結(jié)果。許多最新的人臉跟蹤的研究都遇到了最在背景噪音的問題，且

6、研究都傾向于跟蹤未經(jīng)證實的人臉，例如臂和手。</p><p>　　在本文中，我們提出了一種基于PCA的實時人臉檢測和跟蹤方法，該方法是利用一個如圖1所示的活動攝像機(jī)來檢測和識別人臉的。這種方法由兩大步驟構(gòu)</p><p>　　成：人臉檢測和人臉跟蹤。利用兩副連續(xù)的幀，首先檢驗人臉的候選區(qū)域，并利用PCA技術(shù)來判定真正的人臉區(qū)域。然后，利用特征技術(shù)（eigen－technique）</

7、p><p><b>　　跟蹤被證實的人臉。</b></p><p><b>　　2.人臉檢測</b></p><p>　　在這一部分中，將介紹本文提及到的方法中的用于檢測人臉的技術(shù)。為了改進(jìn)人臉檢測的精確性，我們把諸如膚色模型[1,6]和PCA[7,8]這些已經(jīng)發(fā)表的技術(shù)結(jié)合起來。</p><p>&l

8、t;b>　　2.1膚色分類</b></p><p>　　檢測膚色像素提供了一種檢測和跟蹤人臉的可靠方法。因為通過許多視頻攝像機(jī)得到的一幅RGB圖像不僅包含色彩還包含亮度，所以這個色彩空間不是檢測膚色像素[1,6]的最佳色彩圖像。通過亮度區(qū)分一個彩色像素的三個成分，可以移動亮度。人臉的色彩分布是在一個小的彩色的色彩空間中成群的，且可以通過一個2維的高斯分部來近似。因此，通過一個2維高斯模型可以近似

9、這個膚色模型，其中平均值和變化如下：</p><p>　　m=(,) 其中＝，＝ （1）</p><p>　?。?（2）</p><p>　　一旦建好了膚色模型，一個定位人臉的簡單方法是匹配輸入圖像來尋找圖像中人臉的色彩群。原始圖像的每一個像素被轉(zhuǎn)變?yōu)椴噬纳士臻g，然后

10、與該膚色模型的分布比較。</p><p><b>　　2.2動作檢測</b></p><p>　　雖然膚色在特征的應(yīng)用種非常廣泛，但是當(dāng)膚色同時出現(xiàn)在背景區(qū)域和人的皮膚區(qū)域時，膚色就不適合于人臉檢測了。利用動作信息可以有效地去除這個缺點。為了精確，在膚色分類后，僅考慮包含動作的膚色區(qū)域。結(jié)果，結(jié)合膚色模型的動作信息導(dǎo)出了一幅包含情景（人臉區(qū)域）和背景（非人臉區(qū)域）的二

11、進(jìn)制圖像。這幅二進(jìn)制圖像定義為 ，其中It(x,y)</p><p>　　和It-1(x,y)分別是當(dāng)前幀和前面那幀中像素（x,y）的亮度。St是當(dāng)前幀中膚色像素的集合，（斯坦）t是利用適當(dāng)?shù)拈撓藜夹g(shù)計算出的閾限值[9]。作為一個加速處理的過程，我們利用形態(tài)學(xué)（上）的操作（morpholoical operations）和連接成分分析，簡化了圖像Mt。</p><p>　　2.3利用PC

12、A檢驗人臉</p><p>　　因為有許多移動的對象，所以按序跟蹤人臉的主要部分是很困難的。此外，還需要檢驗這個移動的對象是人臉還是非人臉。我們使用特征空間中候選區(qū)域的分量向量來為人臉檢驗問題服務(wù)。為了減少該特征空間的維度，我們把N維的候選人臉圖像投影到較低維度的特征空間，我們稱之為特征空間或人臉空間[7,8]。在特征空間中，每個特征說明了人臉圖像中不同的變化。</p><p>　　為了簡

13、述這個特征空間，假設(shè)一個圖像集合I1，I2，I3，…，IM，其中每幅圖像是一個N維的列向量，并以此構(gòu)成人臉空間。這個訓(xùn)練（測試）集的平均值用A＝來定義。用i＝I I－A來計算每一維的零平均數(shù)，并以此構(gòu)成一個新的向量。為了計算M的直交向量，其中該向量是用來最佳地描述人臉圖像地分布，首先，使用C＝iir＝Y(jié)Yr （4）來計算協(xié)方差矩陣Y＝[ 1 2… M]。雖然矩陣C是N×N維的，但是定義一個N維的特征向量和N個特征值是個難處理的

14、問題。因此，為了計算的可行性，與其為C找出特征向量，不如我們計算[YTY]中M個特征向量vk和特征值k，所以用u k＝來計算一個基本集合，其中k＝1，…，M。關(guān)于這M個特征向量，選定M個重要的特征向量當(dāng)作它們的相應(yīng)的最大特征值。對于M個訓(xùn)練（測試）人臉圖像，特征向量W i＝[w 1，w 2，…，w M’]用w k＝u kTi,k=1,…，M（6）來計算。</p><p>　　為了檢驗候選的人臉區(qū)域是否是真正的人臉

15、圖像，也會利用公式（6）把這個候選人臉區(qū)域投影到訓(xùn)練（測試）特征空間中。投影區(qū)域的檢驗是利用人臉類和非人臉類的檢測區(qū)域內(nèi)的最小距離，通過公式（7）來實現(xiàn)的。Min（||Wkcandidate－Wface||,||Wkcandidate－Wnonface||），（7）其中Wkcandidate是訓(xùn)練（測試）特征空間中對k個候選人臉區(qū)域，且Wface，Wnonface分別是訓(xùn)練(測試)特征空間中人臉類和非人臉類的中心坐標(biāo)，而||×

16、||表示特征空間中的歐幾里德距離（Euclidean）</p><p><b>　　3.人臉跟蹤</b></p><p>　　在最新的人臉檢測中，通過在特征空間中使用一個距離度量標(biāo)準(zhǔn)來定義圖像序列中下一幅圖像中被跟蹤的人臉。為了跟蹤人臉，位于被跟蹤人臉的特征向量和K個最近被檢測的人臉之間的歐幾里德距離是用obj＝argkmin||Wold－Wk||，k＝1，…，K，（

17、8）來計算的。</p><p>　　在定義了人臉區(qū)域后，位于被檢測人臉區(qū)域的中心和屏幕中心之間的距離用distt（face，screen）＝Facet（x，y）－Screen（height/2，width/2），（9）來計算，其中Facet（x，y）是時間t內(nèi)被檢測人臉區(qū)域的中心，Screen（height/2，width/2）是屏幕的中心區(qū)域。使用這個距離向量，就能控制攝像機(jī)中定位和平衡/傾斜的持續(xù)時間。攝像機(jī)

18、控制器是在這樣的方式下工作的：通過控制活動攝像機(jī)的平和/傾斜平臺把被檢測的人臉區(qū)域保持在屏幕的中央。在表2自己品母國。參數(shù)表示的是活動攝像機(jī)的控制。用偽代碼來表示平衡/傾斜處理的持續(xù)時間和攝像機(jī)的定位。</p><p>　　計算平和/傾斜持續(xù)時間和定位的偽代碼：</p><p>　　Procedure Duration（x，y）</p><p><b>

19、　　Begin</b></p><p>　　Sigd=None；</p><p>　　Distance=；</p><p>　　IF distance> then</p><p>　　Sigd=Close；</p><p>　　ELSEIF distance> then</p&g

20、t;<p><b>　　Sigd=fat；</b></p><p>　　Return（Sigd）；</p><p>　　End Duration；</p><p>　　Procedure Orientation（x，y）</p><p><b>　　Begin</b></p>

21、;<p>　　Sigo=None；</p><p>　　IF x> then</p><p>　　Add “RIGHT” to Sigo；</p><p>　　ELSEIF x<- then</p><p>　　Add “LEFT” to Sigo；</p><p>　　IF y>

22、then</p><p>　　Add “up”to Sigo；</p><p>　　ElSEIF x<- then</p><p>　　Add “DOWN” to Sigo；</p><p>　　Return（Sigo）；</p><p>　　End Orientation；</p><p&

23、gt;<b>　　4.結(jié)論</b></p><p>　　本文中提議了一種基于PAC的實時人臉檢測和跟蹤方法。被提議的這種方法是實時進(jìn)行的，且執(zhí)行的過程分為兩大部分：人臉識別和人臉跟蹤。在一個視頻輸入流中，首先，我們利用注入色彩、動作信息和PCA這類提示來檢測人臉區(qū)域，然后，用這樣的方式跟蹤人臉：即通過一個安裝了平衡/請求平臺的活動攝像機(jī)把被檢測的人臉區(qū)域保持在屏幕的中央。未來的工作是我們將進(jìn)

24、一步發(fā)展這種方法，通過從被檢測的人臉區(qū)域種萃取臉部特征來為臉部活動系統(tǒng)服務(wù)。</p><p><b>　　參考文獻(xiàn)</b></p><p>　　[1] Z. Guo, H. Liu, Q. Wang, and J. Yang, “A Fast Algorithm of Face Detection for Driver Monitoring,” In Proceed

25、ings of the Sixth International Conference on Intelligent Systems Design and Applications, vol.2, pp.267 - 271, 2001.</p><p>　　[2] M. Yang, N. Ahuja, “Face Detection and Gesture Recognition for Human-Comput

26、er Interaction,” The International Series in Video Computing , vol.1, Springer, 2001.</p><p>　　[3] Y. Freund and R. E. Schapire, “A Decision-Theoretic Generaliztion of On-Line Learning and an Application to

27、 Boosting,” Journal of Computer and System Sciences, no. 55, pp. 119-139, 1997.</p><p>　　[4] J. I. Woodfill, G. Gordon, R. Buck, “Tyzx DeepSea High Speed Stereo Vision System,” In Proceedings of the Conferen

28、ce on Computer Vision and Pattern Recognition Workshop, pp.41-45, 2004.</p><p>　　[5] Xilinx Inc., “Virtex-4 Data Sheets: Virtex-4 Family Overview,” Sep. 2008. DOI= http://www.xilinx.com/</p><p>

29、　　[6] Y. Wei, X. Bing, and C. Chareonsak, “FPGA implementation of AdaBoost algorithm for detection of face biometrics,” In Proceedings of IEEE International Workshop Biomedical Circuits and Systems, page S1, 2004.</p&

30、gt;<p>　　[7] M. Yang, Y. Wu, J. Crenshaw, B. Augustine, and R. Mareachen, “Face detection for automatic exposure control in handheld camera,” In Proceedings of IEEE international Conference on Computer Vision Syst

31、em, pp.17, 206.</p><p>　　[8] V. Nair, P. Laprise, and J. Clark, “An FPGA-based people detection system,” EURASIP Journal of Applied Signal Processing, 2005(7), pp. 1047-1061, 2005</p><p>　　[9] C

32、. Gao and S. Lu, “Novel FPGA based Haar classifier face detection algorithm acceleration,” In Proceedings of International Conference on Field Programmable Logic and Applications, 2008.</p><p><b>　　外文原

33、文一</b></p><p>　　PCA-Base Real-Time Face Detection and Tracking</p><p>　　【Abstract】：</p><p>　　This article put forward complicated background term next; realize solid contempor

34、aries face examination with on the trail of a kind of method. These kinds of method regard main composition analysis technique as basal. Facial examination in person for realizing, first, we want to use a skin color mode

35、l to act the information with the some (such as: Posture, signal, expression of eyes).Then, the usage PAC technique examines these drive the district that examine, from but judge a real position. Bu</p><p>&

36、lt;b>　　1 preface</b></p><p>　　Seeing the signal of handles many applications, for example owing to the communication can see the telecommunication meeting that turn, for disable and sick person serv

37、ice of the lips reads the system. In up many systems that mention, the facial examination in person drink to follow to see to can't lack necessarily of constitute the part. In this text, involve the some solid of per

38、son a district follows the [1 － 3 ] .By any large, according to follow the angle different, can is divided in to fol</p><p>　　According to the on the trail of that identify is really with the object identifi

39、es technique is basal, but follow the function of the system is the restrict of the efficiency to suffer to identify the method. According to the on the trail of of the action is a method to depend on to examine the tech

40、nique in the action, and that technique can be been divided in to see flow( optical flow) with the method that act the — energy( motion － energy).</p><p>　　According to the method of the edge useds for the e

41、dge that follow a picture </p><p>　　preface row, but these edgeses is usually the boundary line of the main object.However, because were musted shine on with the light at the color by the on the trail of obj

42、ect the term descends to display the obvious edge changes, so these methodses will fall among the color with the variety that light shine on.In addition, be a background of picture contain very obvious edge,( follow the

43、method) dependable result in very difficult offering.Current this type of method that a lot of cultural herit</p><p>　　In this text, we put forward a kind of according to PCA solid contemporaries an examinat

44、ion with follow the method, that method is an activity to make use of a,such as figure,1 show resemble machine to examine with identify the person facial.This kind of method from two greatest steps composing:Person an e

45、xamination with person's face follow.Make use of two continuouses, examine a person's face candidate for election districts first, combine exploitation PCA technique to judge the real person a</p><p>

46、;　　2 Person an examination</p><p>　　In this first part, will introduce the method that this text mention inside of used for the technique that examine person's face.For improves an accurate for examining

47、, we announce such as the skin color model [ 1,6 ] with PCA [ 7,8 ] these already of the technique knot puts together.</p><p>　　2.1 skin color classification</p><p>　　The examination skin color

48、pixel provides a kind of examination with follow the facial and dependable method in person.Because pass many that sees the machine resemble a RGB picture not only include color but also gets bright degree in containment

49、, so this color space is not the best color to examine the skin color pixel [ 1,6 ] picture.Pass bright a three compositions for distinguishing analyse a color pixel, can move bright degree.A Gauss for of color distribut

50、ing is in a small chromatic color</p><p>　　m=(,) 其中＝，＝ （1）</p><p>　?。?（2）</p><p>　　Once set up to like the skin color model, a positi

51、ons facial and simple method in person is match the importation picture to look for facial color in middleman in picture cluster.Each a pixel of the primitive picture were changed into the chromatic color space, then dis

52、tributing with the skin color's model the comparison.</p><p>　　2.2 action examination</p><p>　　Although the skin color application in characteristic grows very extensive, when the skin color

53、 appear at the same time in the background district with the person's skin district, skin color is not suitable for in the person an examination.Making use of to act information can away with this weakness availably

54、.For the sake of the precision, after the skin color divides into section, consider the skin color district of the containment action only.Result, the action information of the combination </p><p>　　With the

55、 It-1( x, y) respectively is a bright degree for with front an inside pixel( x, y).The St is a current an inside skin color pixel to gather, the t is a worth [ in limit in to makes use of appropriate limit technique comp

56、ute 9 ] .The acceleration that be used as a process handles, we make use of the operation( morpholoical operations) that appearance learn( top) with link the composition analyzes, simplifying the picture Mt.</p>&

57、lt;p>　　2.3 make use of the PCA examine person's face</p><p>　　There is many ambulatory objects, so follow in sequence the facial and main part in person is very difficult.In addition, return the deman

58、d examine this ambulatory object is person's face or not person's face.We uses characteristic space inside the weight vector of the candidate for election district to behave face examination problem service.For r

59、educing that characteristic the spatial a candidate for, we N a picture casts shadow the characteristic space of the lower the degree of , we call it </p><p>　　Say this characteristic space for the sake of C

60、hien, suppose a picture gather the I1, I2, I3, … , IM, among them each picture is the row vector of a N , and with this composing person a space.The average value that this training( test) gather uses the A= define.Use t

61、he i= the I I － A computes the zero average number of each , and with this composing a new vector.For computing the M keep handing over vector, among them that vector is to uses to come to describe the person best a pict

62、ure ground di</p><p>　　For the sake of the person of the examination candidate for election whether a district is a real person or not a picture, also will make use of the formula(6) cast shadow the training

63、( test) characteristic space inside to this candidate a district.Examination that cast shadow the district is a minimum distance to makes use of a person's face with not person's face examination district inside,

64、 passing the formula(7) come to something to realizes.Min(|| Wkcandidate － Wface||,|| Wkcandidate － Wnon</p><p>　　3.Person's face follows</p><p>　　In latest person an examination, pass to us

65、e a distance generous character standard to define the picture preface row in characteristic space inside a picture inside drive on the trail of person's face.For following a person's face, locate to is followed

66、a person's face the characteristic vector is recent to is examined with the of K of person the of the an is several in the virtuous distance is to uses the obj= argkmin|| Wold － Wk||, k=1, … , K,(8) compute of.</p

67、><p>　　After defining the person a district, locate to is examined person the center of a district with distance that hold the act center uses the distt( face, screen)= Facet( x, y) － Screen( height/2, width/2)

68、,(9) compute, among them Facet( x, y)</p><p>　　The that time a t inside were examined the person the center of a district, the Screen( height/2, width/2) is a center to hold the act district.Use this distanc

69、e vector, can control the resemble to position in the machine with equilibrium/ tilt to one side of continuously time.The resembles the machine controller is what under such way work:Pass to control the activity resemble

70、 the machine even with/ tilt to one side the terrace examines drive of person a district keeps at hold the act central.I</p><p>　　The calculation is even with/ tilt to one side keep on time with the false co

71、de that position:</p><p>　　Procedure Duration（x，y）</p><p><b>　　Begin</b></p><p>　　Sigd=None；</p><p>　　Distance=；</p><p>　　IF distance> t

72、hen</p><p>　　Sigd=Close；</p><p>　　ELSEIF distance> then</p><p><b>　　Sigd=fat；</b></p><p>　　Return（Sigd）；</p><p>　　End Duration；</p>

73、<p>　　Procedure Orientation（x，y）</p><p><b>　　Begin</b></p><p>　　Sigo=None；</p><p>　　IF x> then</p><p>　　Add “RIGHT” to Sigo；</p><p>　

74、　ELSEIF x<- then</p><p>　　Add “LEFT” to Sigo；</p><p>　　IF y> then</p><p>　　Add “up”to Sigo；</p><p>　　ElSEIF x<- then</p><p>　　Add “DOWN” to Sigo

75、；</p><p>　　Return（Sigo）；</p><p>　　End Orientation；</p><p>　　4.Conclusion</p><p>　　It suggested in this text a kind of according to PAC solid contemporaries face examina

76、tion with follow method.Were been a solid hour to proceed by this kind of method that suggest of, and the executive process is divided into two big part:Person's face identifies to follow with person's face.In fi

77、rst saw input flow, first, we make use of the infusion color, action the information is this type of to hint to examine the person a district with the PCA, then, use such way follow person's face:Passed</p>&l

78、t;p>　　REFERENCES</p><p>　　[1] Z. Guo, H. Liu, Q. Wang, and J. Yang, “A Fast Algorithm of Face Detection for Driver Monitoring,” In Proceedings of the Sixth International Conference on Intelligent Systems

79、 Design and Applications, vol.2, pp.267 - 271, 2001.</p><p>　　[2] M. Yang, N. Ahuja, “Face Detection and Gesture Recognition for Human-Computer Interaction,” The International Series in Video Computing , vo

80、l.1, Springer, 2001.</p><p>　　[3] Y. Freund and R. E. Schapire, “A Decision-Theoretic Generaliztion of On-Line Learning and an Application to Boosting,” Journal of Computer and System Sciences, no. 55, pp.

81、119-139, 1997.</p><p>　　[4] J. I. Woodfill, G. Gordon, R. Buck, “Tyzx DeepSea High Speed Stereo Vision System,” In Proceedings of the Conference on Computer Vision and Pattern Recognition Workshop, pp.41-45,

82、 2004.</p><p>　　[5] Xilinx Inc., “Virtex-4 Data Sheets: Virtex-4 Family Overview,” Sep. 2008. DOI= http://www.xilinx.com/</p><p>　　[6] Y. Wei, X. Bing, and C. Chareonsak, “FPGA implementation of

83、 AdaBoost algorithm for detection of face biometrics,” In Proceedings of IEEE International Workshop Biomedical Circuits and Systems, page S1, 2004.</p><p>　　[7] M. Yang, Y. Wu, J. Crenshaw, B. Augustine, an

84、d R. Mareachen, “Face detection for automatic exposure control in handheld camera,” In Proceedings of IEEE international Conference on Computer Vision System, pp.17, 206.</p><p>　　[8] V. Nair, P. Laprise, an

85、d J. Clark, “An FPGA-based people detection system,” EURASIP Journal of Applied Signal Processing, 2005(7), pp. 1047-1061, 2005</p><p>　　[9] C. Gao and S. Lu, “Novel FPGA based Haar classifier face detection