CCD雙目立體視覺測量系統的理論研究外文翻譯

1、畢業(yè)設計外文資料翻譯題 目: ccd stereo vision measurement system theory 院系名稱：電氣工程學院 專業(yè)班級： 自動化 學生姓名： 學 號： 指導教師： 教師職稱： 起止日期：2011-2-262011-3-14 地點: 附 件： 1.外文資料翻譯譯文；2.外文原文。 指導教師評語： 簽名： 年 月 日附件1：外文資料翻譯譯文ccd雙目立體視覺測量系統的理論研究摘要: 利用幾何成像原理建立起 ccd 雙目立體視覺測量系統的數學模型 ,從提高系統測量精度出發(fā) ,在理論上重點對系統結構參數、 圖像識別誤差與系統測量精度的關系進行了深入的分析和探討 ,并通過

2、實驗對結論進行了驗證。研究內容對實際建立該測量系統具有很強的指導作用。關鍵詞: 立體視覺; ccd ; 測量精度; 圖像識別; 系統測量引言雙目立體視覺測量技術是計算機視覺中的一個重要分支,一直是計算機視覺研究的重點和熱點之一。由于其近似于人眼視覺系統 ,具有較高的測量精度和速度 ,并具有結構簡單 ,便于使用等優(yōu)點 ,所以被廣泛應用于工業(yè)檢測、 物體識別、 工件定位、 機器人自導引等諸多領域。近年來許多學者對此進行了大量的研究工作1 - 4 。其中大量的工作集中在對視覺測量系統的數學模型、 系統的定標方法5 - 7 以及目標特征點匹配算法8 - 9 的研究上 ,而對系統的結構參數(兩個 ccd

3、之間的距離、 光軸夾角等)研究得卻很少。文獻10 對立體視覺結構參數進行了相應的理論研究 ,但它是從觀看物體時的深度感出發(fā)研究ccd與物體之間的距離、 兩個ccd間距和觀看距離3個參數之間的關系 ,沒有涉及到結構參數對系統測量精度的影響。而實踐證明系統的結構參數設置在實際應用中對于系統的測量精度是至關重要的。此外 ,從立體視覺測量原理中 ,可以看出圖像識別誤差是另一個對系統測量精度產生直接影響的重要因素。綜合以上考慮 ,從理論上對系統的結構參數設置和圖像識別誤差對系統測量精度的影響進行了深入的分析和研究。結合系統結構參數對攝像機定標精度的影響 ,給出了實際應用中組建雙目立體視覺測量系統的設計方

4、案。1 雙目立體視覺測量原理及數學模型1. 1 攝像機成像模型攝像機的成像模型,是光學成像系統幾何關系的數學表示。目前在攝像機標定中應用的攝像機成像模型主要有針孔成像模型、 雙平面模型和人工神經網絡模型等。其中針孔成像模型是目前大量采用的一種成像模型,它反映的是一種理想的線性映射關系,如圖1 所示。其中, oc為攝像機的光心, oc xcyc zc為攝像機坐標系, oxyz 為世界坐標系, oxy 為攝像機成像平面物理坐標系。p( x , y , z) 為空間一物點, p ( x , y)是其在圖像平面上的投影點。根據該成像模型,空間物點 p( x , y , z)與像點之間的關系可表示為當兩

5、個或兩個以上攝像機進行交會時,可以得到2 i ( i 2)個方程所組成的超定方程組,因此可以用最小二乘法對方程組求解以確定空間物點的坐標。1. 2 雙目立體視覺測量系統的數學模型雙目立體視覺系統通常由兩臺結構和性能完全相同的ccd組成 ,并且兩個ccd擺放位置對稱?；谏鲜鰯z像機成像模型 ,由式(2)可以推出如下超定方程組:其中 r , t , r , t 分別為兩攝像機坐標系相對于世界坐標系的平移和旋轉矩陣,可以通過攝像機標定得到。( x , y) , ( x , y )分別為空間物點( x , y , z)在兩ccd圖像平面的投影的物理坐標。圖2 雙目立體視覺簡易模型為了對系統結構參數進行

6、分析,這里構建如圖2所示的立體視覺系統。在該結構中,三坐標系處于同一平面內,其中 y 軸垂直紙面向里,世界坐標系 oxyz 與左攝像機坐標系oc xc yc zc原點重合,兩攝像機間距為 l ,兩光軸夾角為 2。根據上述結構,可以確定兩攝像機組成的超定方程組為則由式(4)可得: 2 圖像識別誤差對系統測量精度的影響立體視覺系統中 ,空間物點在兩個攝像機圖像平面上的位置是通過像素坐標來表示的 ,而面陣ccd攝像機像素具有一定的物理尺寸 ,這就使得空間物點在圖像上的真實物理坐標無法得到準確的表達 ,從根本上造成了圖像識別誤差。如圖3所示。圖像平面上像素坐標系與物理坐標系有如下關系:其中: ( u

7、, v)是以像素為單位的圖像坐標系的坐標,dx , dy 分別為每一個像素在x 軸和 y 軸方向上的物理尺寸, ( u0 , v0)為 oxy 坐標系原點o在o0 uv 坐標系中的坐標3 。如圖 4 所示,假設圖像識別精度達到0. 5個像素級。對于一空間物點,設其投影到圖像平面上的第 i 行,第 j 列的像素中,則此時該點的物理坐標為 x = ( i - u0) dxy = ( j - v0) dy即只要該點落在該像素內,其坐標值是一個定值。而理想情況下的坐標應分別在一定的范圍內:( i - u0 - 0. 5) dx x ( i - u0 + 0. 5) dx( j - v0 - 0. 5)

8、 dy y ( i - v0 + 0. 5) dy綜合上式,則對應該點的圖像識別誤差為 ex = 0. 5 - ( x - x )ey = 0. 5 - ( y - y )其中: x , y 分別表示對 x , y 取整。由于圖像識別誤差的存在,則實際像點坐標與理想像點坐標有如下關系: x = x + ex , y = y + ey , ( x , y)為實際像點物理坐標, ( x , y )為理想像點物理坐標。設被測物點坐標 p = ( x , y , z) ,則由公式(1)計算出該點在兩攝像機投影的理想像點坐標( x , y)和( x , y ) ,考慮圖像識別誤差,根據式(8)得出實際像

9、點坐標( x ,y) , ( x , y ) ,將其代入(4)式可得被測點的空間坐標( x , y , z) ,則被測物點的測量誤差可表示為ex = x - x , ey = y - y , ez = z - z (9)3 系統結構參數分析及實驗結果3. 1 結構參數與系統測量精度的關系根據上述系統模型及數學推導過程,得出了系統結構參數與系統測量誤差之間的關系圖。圖5為特定物點的測量誤差與兩攝像機夾角2之間的關系圖。從圖中可以看出, ex 變化波動不大, ey 隨 2的增大呈緩慢的上升趨勢, ez 則變化比較劇烈,隨2的增大大幅提升。圖5 特定空間點測量誤差與兩軸夾角的關系圖圖6為一定范圍內的

10、物點與世界坐標系原點的平均測量誤差與兩攝像機夾角 2和兩攝像機間距離l 之間的關系圖?？梢钥闯?當兩攝像機間距離一定時,平均誤差隨 2的增大而增大,當兩攝像機夾角一定時,平均誤差隨其距離的增加而不斷增大。圖6 平均誤差與兩軸夾角、 兩攝像機距離的關系圖總體上講,結構參數與系統測量精度是一個較為復雜的函數關系,可以總結如下:1) 攝像機之間的間距大小與系統測量誤差成正比關系,間距越小,誤差越小;2) 當攝像機之間的夾角不大于130 時,測量誤差較小,反之較大。有極高的可信性。3. 2 結構參數對攝像機定標精度的影響在實際應用中,需要建立兩個 ccd 坐標系之間的聯系 ,這就需要對兩 ccd 進行

11、立體定標 ,以求取兩 ccd之間的旋轉矩陣和平移矩陣。事實證明攝像機定標精度與系統結構參數設置同樣有著非常緊密的關系。定標方法采用的是基于單平面模板定標策略 ,精度評估采用基于棋盤格長度的評估方法。該方法亦可以作為系統測量誤差的測量方法。對定標模板上的10個長為 50 mm的棋盤格進行了反復實驗 ,實驗結果見表1。表1 結構參數對系統定標平均誤差的影響從表1中可以看出 ,當攝像機距離一定時 ,定標誤差隨光軸夾角的增加而不斷增加 ,當光軸夾角固定不變時 ,定標誤差隨攝像機距離的增加而不斷增加。大量實驗證明 ,當兩攝像機間距離不超過 500mm ,兩光軸夾角不超過60 時 ,定標誤差較小。4 結束

12、語綜合系統結構參數對測量精度及定標精度的影響 ,在建立立體視覺系統時 ,兩攝像機光軸夾角和兩攝像機間距應盡可能小 ,但在實際應用中 ,考慮到便于目標特征點視覺匹配 ,尤其是對運動目標進行大范圍實時跟蹤測量 ,有目標被遮擋的情況發(fā)生時 ,兩攝像機光軸夾角應選擇在 30 60 之間。此外 ,目標特征點圖像識別精度應盡可能達到亞像素精度 ,盡量避免 “失之毫厘 ,差之千里” 的現象。根據以上推導搭建了立體視覺系統,取得了較好的實驗結果,證明本文的結論對實踐具有較大的指導意義,為進一步開展深入研究打下了堅實的基礎。附件2：外文原文（復印件）ccd stereo vision measurement s

13、ystem theoryabstract: using the principles of geometrical imaging ccd stereo vision to build a mathematical model of measurement systems, improve the precision of the system from the start, in theory, focus on system parameters, image recognition errors and the relationship between the precision of

14、the system has in-depth analysis and explore and experiment on the conclusion was verified. research on the actual establishment of this system has a strong guiding role.key words: stereo vision; ccd; measurement accuracy; image recognition; system measurementsintroduction stereoscopic measurement i

15、n computer vision technology is an important branch of computer vision has been the focus of the study and the hot spots. due to its similar to the human visual system with high precision and speed, and has a simple structure, easy to use, etc., they were widely used in industrial inspection, object

16、 recognition, workpiece positioning, robot homing, and many other fields . in recent years many scholars have done a lot of research work 1 - 4. a lot of work which focused on the mathematical model of vision measurement system, the system calibration method 5 - 7 and the target feature point matchi

17、ng algorithm 8 - 9, while the structural parameters of the system (of two ccd distance between the optical axis angle, etc.) are rarely studied. 10 structure parameters of stereoscopic vision corresponding theoretical research, but it is viewing objects from the start of the depth of feeling between

18、 the ccd and the object distance and viewing distance of the two ccd spacing between 3 parameters , did not address the structural parameters of the system measurement accuracy. the practice proved that the structure of the system parameters in practical applications of the systems measurement accur

19、acy is essential. in addition, from the stereo vision measurement principle, we can see the error image recognition system accuracy is another important factor in a direct impact. based on the above considerations, the structure from the theoretical parameters of the system and image recognition err

20、ors on the precision of the system of in-depth analysis and research. combination of system parameters on the accuracy of the camera calibration given set of practical applications, stereo vision measurement system design1 binocular stereo vision measurement principle and mathematical model1.1 camer

21、a imaging modelcamera imaging model, the geometric relationship between the optical imaging system of mathematics said. at present the application of camera calibration are pinhole camera imaging model imaging model, two-plane model and artificial neural network model. in which a large number of pin

22、hole imaging model is used in an imaging model, which reflects the relationship between an ideal linear map shown in figure 1. one, oc for the camera optical center, oc xcyc zc for the camera coordinate system, oxyz the world coordinate system, oxy for the camera imaging plane physical coordinates.p

23、 (x, y, z) for the space of a material point, p (x, y) is its projection point on the image plane. according to the imaging model, the space object point p (x, y, z) and the relationship between the image point can be expressed aswhen two or more intersection cameras, you can get 2 i (i 2) consistin

24、g of equations overdetermined equations, so the least squares method can be used to solve equations to determine the spatial coordinates of object points.1.2 stereo vision measurement system modelbinocular stereo vision system usually consists of two identical ccd structure and properties of the com

25、position and placement of the two ccd symmetry. based on the above camera imaging model, from (2) can introduce the following overdeterminedequations:where r, t, r , t are the two camera coordinate system relative to the world coordinate system translation and rotation matrix, can be obtained throug

26、h camera calibration. (x, y), (x , y), respectively, for the space object point (x, y, z) in the two ccd image plane projection of the physical coordinates.figure 2 simple model of binocular stereo visionin order to analyze structural parameters of the system, where the building shown in figure 2 st

27、ereo vision system. in this structure, the coordinate system in the same plane, which for the y axis perpendicular to the paper, the world coordinate system oxyz with the left camera coordinate origin coincides oc xc yc zc, two camera spacing l, the angle between the two axis 2. according to the abo

28、ve structure, the camera can determine the composition of the two equations for the overdeterminedby equation (4) yields:2 image recognition errors on the accuracy of the systemstereo vision system, the space object point in two camera image plane position is represented by pixel coordinates, and pi

29、xel area array ccd camera has a certain physical size, which makes the space object point in the image on the real physical coordinates not the exact expression, a fundamental cause of the image recognition errors.figure 3.image plane pixel coordinates and physical coordinates the following relation

30、s: where: (u, v) the image in pixels coordinate system, dx, dy, respectively, for each pixel in the x-axis and y axis on the physical size, (u0, v0) o the origin of the coordinate system for the oxy o0 uv coordinate system in the coordinates 3. shown in figure 4, assuming that the image recognition

31、accuracy of 0.5 pixel. for a space object point, set the projection to the image plane on the i-line, the first j columns of pixels, then this time the physical coordinates of the point x = ( i - u0) dxy = ( j - v0) dy as long as the point falls within the pixel, the coordinate value is a constant.

32、ideally, the coordinates of which should be within a certain range, respectively:( i - u0 - 0. 5) dx x ( i - u0 + 0. 5) dx( j - v0 - 0. 5) dy y ( i - v0 + 0. 5) dyintegrated on the type, then the error should point to image recognition ex = 0. 5 - ( x - x )ey = 0. 5 - ( y - y )where: x, y, respectiv

33、ely, for x, y rounded. since the existence of image recognition errors, the actual coordinates of image point coordinates and the ideal image point has the following relationship: x =? x + ex, y =? y + ey, (x, y) as the point of actual physical coordinates, ( ? x,? y) as the ideal image point of phy

34、sical coordinates. let the measured object point coordinates p = ( x, y, z), by equation (1) to calculate the projection of the points in the two cameras the ideal image point coordinates (? x,? y) and (? ? x ,? y), consider the image recognition error, according to equation (8) come to the actual p

35、ixel coordinates (x, y), (x , y), be substituted into (4), we have been space coordinates of measuring point (x, y, z), the measurement error of the measured object point can be expressed as ex = x - x, ey = y - y, ez = z - z (9)3 system parameters analysis and experimental results3. 1 structural pa

36、rameters of the relationship with the precision of the system and mathematical model based on the above derivation process, come to the system structure and system parameters of the relationship between measurement error map. figure 5 is a specific object point measurement error and the camera angle

37、 2 between the two diagrams. it can be seen from the figure, ex volatility is not changing, ey 2 increases with the rising trend was slow, ez is more violent change, with the increase of 2 increased significantly.figure 5 the measurement error is a specific spatial point the angle between the two ax

38、is graphfigure 6 for a certain object point within the coordinate origin with the world average measurement error and the two cameras and two camera angle 2 between the distance between the l map. it can be seen, when a certain distance between the two cameras, the average error with the increase of

39、 2, when the angle between the two cameras is fixed, the average error increases with the distance increasing.figure 6 average error and the angle between two axes, the distance between the two cameras chartoverall, structural parameters and precision of the system is a more complex function can be

40、summarized as follows:1) spacing between the camera and the measurement error is directly proportional to the smaller distance, the smaller the error;2) when the camera is not greater than the angle between the 130 , the measurement error is small, whereas larger. has high credibility.3. 2 structura

41、l parameters on the accuracy of camera calibrationin practice, the need to create two links between ccd coordinates, which need to be two-dimensional ccd calibration, in order to obtain the rotation between the two ccd matrix and translation matrix. proved the precision of camera calibration paramet

42、ers and system structure also has a very close relationship. calibration method uses a template based on single-plane calibration strategy, the accuracy was assessed using an assessment based on the length of the checkerboard method. the method also can be used as measurement error of the measurement method. calibration template of 10 to 50 mm length of the checkerboard was repeated expe