Stereoscopic Particle Image Velocimetry using telecentric lenses yx T. Fournel y y y , S. Coudert , C. Fournier , and C. Ducottet Laboratoire Traitement du Signal et Instrumentation, UMR CNRS 5516 Universite J. Monnet 23, rue du Dr P. Michelon, 42023 Saint-Etienne cedex 2, France y Telecentric imaging is presented in the general frame of Digital Stereoscopic Particle Image Velocimetry (DSPIV). A 2D-3D mapping approach is combined with telecentric imaging in order to obtain a user-friendly DSPIV technique. This technique is tested on translation tests then in a round free jet. Abstract. Stereoscopic Particle Image Velocimetry, Telecentric lens, AÆne model, Mapping function Keywords: PACS numbers: 06.30.G, 07.05.P Submitted to: Meas. Sci. Technol. x To whom correspondence should be addressed (fournel@univ-st-etienne.fr) 2 SPIV using telecentric lenses 1. Introduction Thanks to digital techniques, Particle Image Velocimetry allows the user-friendly measurement of the velocity eld of a uid ow in a laser sheet. In order to get the three components of the displacement between two exposures, video-cameras are positioned in stereoscopic conguration. Two congurations of imaging system have been successfully developed for DSPIV (Digital Stereoscopic Particle Image Velocimetry) : the translation method (Solo et al 1997, Lecerf et al 1999) and the angular displacement method (Willert et al 1997, Oord 1997) depending on whether the optical axes are at right angle to the object plane or not. From a theoretical point of view, the angular displacement method presents a higher sensitivity whereas the maximum possible viewing angle in the translation method is limited by the o-axis aberrations and the decrease of the modulation transfer function toward the edges of the eld-ofview. As conventional lenses are commonly used (Prasad 2000), perspective projections induce image deformations due to the oblique orientation of the object plane. Such deformations resulting from spatially varying magnication are increased when videocameras are focused on the object plane using Scheimpug condition (Hinsch et al 1993, Prasad and Jensen 1995). Then, magnication at a given location in the object plane diers from left to right images. An approach consists in dewarping images or vectors before reconstructing the three components of the displacement eld (Oord 1997). Another approach (Solo et al 1997) makes use of a mapping function F between a 3D location in the object plane and the corresponding 2D location in the left (respectively the right) image plane. To rst order, the displacement is related to the displacement on a given image plane by = rF ( ) (1) In camera coordinate system, the jacobian matrix rF at location , also termed the magnication matrix, is given (Solo et al 1997) by ! 1 0 tg (v ) rF ( ) = M (z) 0 1 tg( ) (2) v where v and v are the angles of the line of sight with respect to the optical axis z. For simplicity, equation (2) is expressed as the optical axis of the imaging system is at right angle to the object plane. A telecentric imaging system can be used to avoid such deformations (Hinsch 1995, Konrath and Schroder 2000, 2002, Fournel et al 2000). Indeed a telecentric lens can perform a parallel projection in the direction of the optical axis. Thus, the angle v and v are null and magnication M is unvarying. Consequently no perspective eect arises. In (Konrath and Schroder 2000) telecentric lenses are used to investigate an air pipe ow. As magnication is unvarying, a dewarping technique (dewarping of vectors) is just performed to compensate optical distortions before reconstruction. In order to make Stereoscopic PIV more and more user-friendly, a 2D-3D mapping approach is here combined with telecentric imaging. The mapping function, given by the X x X x x X X X X 3 sum of the geometrical image and distortions (section 3), is derived from section 2 where telecentric designs are discussed in the frame of SPIV. The displacement equation is then deduced. Section 4 is dedicated to experimental results. Measurement uncertainties estimated by shifting a particle-seeded block of resin are rst discussed. Afterwards SPIV measurements carried out in a round free jet are compared to experimental results given in literature. SPIV using telecentric lenses 2. Imaging with telecentric lenses 2.1. Perfect imaging By positioning an aperture stop in the back focal plane of a lens L1 , the principal rays of light (i.e. the rays passing through the aperture stop) are parallel to the optical axis in object side. Furthermore, despite defocus blurring, the coordinates of the image on the sensor stay unvarying as the object point moves in depth. Such a system whose the entrance pupil is at innity, is said to be telecentric in object side. Obviously the maximum size of the eld of view is limited by the diameter of the lens L1 . A second lens L2 is used to adjust the magnication. Aperture stop (F1’=F2) L1 L2 Image plane x F2’ F1 f1 z f2 Telescopic system. Principal rays (dashed line) are parallel to the optical axis on both object and image sides. Figure 1. The lens L2 can be located one focal length f2 behind the aperture stop in order to have telecentricity in image side (gure 1). Thus, principal rays are parallel to the optical axis in both object and image sides. Such an afocal system, said telescopic, has an unvarying magnication given by the ratio : f (3) M= 2 f1 A more compact system is obtained by positioning the lens L2 in the aperture plane (gure 2). So the image conjugate of a given (out-of-axis) point is the intersection between the principal ray and the ray that should pass (without aperture stop) by the center of the rst lens. However this last ray changes as the object point moves in depth. 4 More exactly magnication depends on the working distance z = zo according to the following relationship M (zo ) = f1 f2 =( f12 + f1 :f2 zo :f2 ) (4) An object-sided telecentric system is usually designed to work at a distance z0 equal to one focal length f1 . The magnication is then also given by equation (3). SPIV using telecentric lenses L1 Image plane x Aperture stop L2 F2’ z F1 z0=f1 f1 f2 System telecentric on object side only. Such a system is more compact than the telescopic one. Principal rays (dashed lines) are no longer parallel to the optical axis on image side. Figure 2. In both cases, the focuses F1 and F2 0 are conjugate. By working at a distance equal to one focal length f1 , the sensor is then located at the back focal plane of the second lens L2 and the image cone of light is determined by the aperture of the lens L2 focused at innity. Consequently the aperture number of such systems is the nominal f-number f2 # of the lens L2 . So, the depth of eld Æz can be deduced from (Adrian 1991) : 2 Æz = 4(1 + M 1 )f2 # (5) where is the wavelength of the laser sheet. 2.2. Scheimpug condition In conventional angular SPIV, the sensor is tilted according to the Scheimpug condition (Hinsch et al 1993, Prasad and Jensen 1995) in order to be focused on the laser sheet and to work with high aperture. The Scheimpug condition is given by tan = M (6) tan where is the viewing angle and the Scheimpug angle (in the horizontal plane). This condition is still valid for a thick lens system as for a telescopic lens system (Konrath and Schroder 2000). The working distance being xed to f1 , it can be easily proved for both telescopic and single-sided congurations by tracing the "ray" coming from the 5 SPIV using telecentric lenses α θ L1 Image plane L2 F1 F2’ f1 Figure 3. f2 Scheimpug condition for telecentric lenses focus F1 (the dashed line in gure 3). It comes out the lens L1 parallel to the optical axis then passes through the focus F2 0 . One can notice from equation (4) that an object-sided telecentric lens system has a varying magnication when tilting the sensor according to the Scheimpug condition. For a 1/2 inch video-camera, the relative magnication can reach 4% for = 45o. Consequently to provide in-focus images at unvarying magnication leads to increase the aperture number f2#. In return the light energy that passes the aperture stop decreases. On the contrary, the magnication of a telescopic lens system remains unvarying in Scheimpug condition because it is telecentric in image side (principal rays are parallel to the optical axis in image side). Its eld of view decreases with a factor cos (whereas the resolution increases with a factor 1=cos) stretching the image with a factor cos=cos. 2.3. Distortions As described in section 2.1, a telecentric lens is a double lens system. So distortions are generated by both lenses L1 and L2 . A slight angle Æ 1 (respectively Æ 2) exists between the actual principal ray and the ideal one in object side (respectively in image side) as illustrated in gure 4. Distortions Æx along x-axis is the sum of distortions introduced by each lens : Æx = B 0 B1 + B1 B12 From equation (3), this last expression can be approximated by Æx ' f2 (Æ 2 Æ 1 ) (7) when viewing at right angle. Distortions can be fully compensated when imaging with a unit-magnication telescopic lens by using a lens L2 identical to the lens L1 (4-f1 set-up). Otherwise, 6 SPIV using telecentric lenses L1 B Aperture stop (F1’=F2) δψ1 Image plane x L2 δψ2 f1 f1 Figure 4. f2 z B12 B1 B’ f2 Distortions of telecentric lenses distortions have to be taken into account. For object-sided telecentric lenses, the angle Æ 2 can be neglected in comparison to Æ 1 because the lens L2 located nearby the aperture works in Gauss conditions. The value of Æ 1 is usually about 0:1Æ. 3. 2D-3D mapping approach 3.1. Video-camera model From section 2, telescopic systems (in Scheimpug condition or not) and compact telecentric ones without tilt have at once an unvarying magnication M and principal rays parallel to the optical axis in object side (v = v = 0). Compared with the expression (2), the magnication matrix of such a system becomes unvarying. To take account for the oblique orientation of the video-camera c = 1; 2 with respect to the object plane, the rotation from the object coordinate system XY Z to the camera coordinate system F20(c) x(c) y(c)z(c) is introduced. Thus, the magnication matrix is expressed by 0 rF (c) = M (c) @ (c) r(c) r(c) 1 r11 12 13 (c) r(c) r(c) A r21 22 23 (8) where rij(c) (i = 1; 2 and j = 1; 2; 3) are the coeÆcients of the rotation matrix. Such a system performs a parallel projection by scaling in F2 0(c) x(c) y(c) the two rst coordinates of an object point given by its coordinates in XY Z : (c) = rF (c) + ! where !(c) is the parallel projection of the origin . This aÆne transformation of coordinates is in practice aected by distortions Æ (c) at the image point x(c) so that the 2D-3D mapping function is expressed by (c) = rF (c) + (c) (9) ! +Æ For a standard sensor format, distortions can be approximated by Æ (c) = a1 r2 (x(c) ; y (c))t (Tsai 1987). They increase when the distance r from the image center (i.e. the intersection between the optical axis and the image plane, here F2 0(c) ) increases. X X x x (c) x x X x (c) x x 7 They are generally about few pixels in the corner of the image. The image grid corresponding to the mesh grid dened in the object plane for eld measurements has to be corrected. The preliminary stage of calibration includes the determination of the magnication matrices and distortions of the left and right video-cameras. SPIV using telecentric lenses 3.2. Displacement equation The relationship between the three-dimensional displacement eld and the displacement eld (c) computed in the image plane can be approximated at the rst order by dierentiating equation (9) : g(c) ' rF (c) (10) where g(c) is the image eld obtained by subtracting the distortion eld dÆ (c) from the image displacement eld (c) . At each position , the three unknowns X , Y , Z can be deduced from equation (10) expressed for the left and right telecentric video-cameras (c = 1; 2) by solving the overdetermined linear system : 0 g (1) 1 0 1 (1) r11 (1) M (1) r12 (1) M (1) r13 (1) 0 x 1 M B C B C (1) X B C B (1) C (1) (1) (1) (1) (1) g B y CC = BB M r21 M r22 M r23 CC BB Y CC B (11) B B @ A (2) r11 (2) M (2) r12 (2) M (2) r13 (2) C gx(2) C B C B C M @ (2) A @ A Z (2) (2) (2) (2) (2) (2) g M r M r M r 21 22 23 y X x x X x x x X 4. Experimental results The stereoscopic set-up used to test the proposed technique has two video-cameras positioned horizontally in symmetrical angular conguration with the viewing angles . Each video-camera is equipped with a compact telecentric lens (Melles & Griot 59LGJ423-LGC535 : f1 = 230mm, M = 0:15) and with a 640 480 pixel 1/2" CCD sensor. Pincushion distortions aect images. The magnitude of distortions obtained after calibration are less than 3 pixels in the corner of the image area in both experiments hereafter reported. They are compensated during the image grid denition. On each image plane, displacement (about 3 pixels) is computed by cross-correlation with the discrete oset technique introduced by Westerweel et al (1997) and with the parabolic three-point estimator. The 3D displacement linearly reconstructed according to equation (11) is slightly distorted by 0:02 pixel per pixel (i.e. for a displacement of one pixel) in the corner of the vector eld. So, distortions have been neglected in the reconstruction of the 3D displacement eld. 4.1. Translation tests A target was accurately shifted in order to study the uncertainties of displacement measurements. The target made in transparent resin (74x92x8 mm3 ) and seeded with 8 SPIV using telecentric lenses silica crystals (Lawson and Wu 1997) was positioned in a laser sheet plane of 2 mm thick. It was step-by-step shifted by means of a remote control XY Z translation stage (resolution < 1m). Displacements were so generated with respect to the coordinate axes and the rst diagonal. The view angle was set to = 30Æ. The size of the eld is 38x25 mm2 . The eld reconstructed at each step has 266 displacement vectors. RMS deviation (µm) 20 16 12 8 4 0 -273 -195 -117 -39 39 117 195 273 ∆ Z (µm) RMS deviations X (triangles), Y (squares) and Z (diamonds) for out-of-plane displacements of a particle-seeded resin lighted with a laser sheet Figure 5. In all the cases, measurements present a bias lower than 3%. Concerning the RMS (Root Mean Square) deviations of the components, they are quite constant over the range of displacement as it is shown on gure 5. This behavior agrees with the theoretical values : "X = "=(M cos ); "Y = "=M; "Z = "=(M sin ) (12) where " denotes the measurement uncertainty on image displacement coordinates. These expressions are derived from 0 1 0 (1) + x(2) )=cos 1 ( x X B C 1 B C B C (13) @ Y C A = 2M B @ (y (1) + y (2) ) A (1) (2) Z (x x )=sin extracted from (11) by rough estimating the experimental set-up with the parameter values : M (i) = M , r11 (i) = cos , r13 (1) = r13 (2) = sin , r12 (i) = 0 = r21 (i) = r23 (i) and r22 (i) = 1 for i = 1; 2. The ratio between the RMS deviations Z and X extends from 1.69 to 1.78 which is close to the theoretical value 1= tan ' 1:73 deduced from equation (12). The ratio between the RMS deviations X and Y is also close to the theoretical value 1= cos ' 1:15. 9 SPIV using telecentric lenses 4.2. Measurements in a round free jet The stereoscopic system was used to investigate a round free jet in a water-lled tank (440x450x900 mm3 ). The Reynolds number based on the nozzle hole diameter of 5 mm is 21000. The ow was seeded with 15 m particles. It was illuminated with a 30 mJ=pulse Nd:Yag laser from the top of the test facilities with a pulse separation of 1.5 ms. In order to minimize aberration eects, the video-cameras were positioned with a normal incidence regarding to the interface (Prasad and Jensen 1995). The view angle was set to = 45Æ. The stereoscopic system pointed out to the axis of the jet, just beyond the developing region where the shear ow is turbulent and consequently three-dimensional. Thus, each line of left and right images is centered with respect to the jet axis and corresponds to a given value of X=d in the 28-35 range. 1 0.9 0.8 <U>/Uo 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0 0 0.02 0.04 0.06 0.08 η 0.1 0.12 0.14 Non-dimensional mean axial velocity hU i=U0 of the round free jet v.s. the non-dimensional radius (Reynolds number of 21000, X=d in the 28-35 range) : t of all the measured proles (continuous line) and theoretical prole (dashed line) Figure 6. As displacement is no longer constant regarding to translation tests (section 4.1), the misalignment between the light sheet coordinate system and the target coordinate system XY Z is accounted for using grid correction techniques (Coudert and Schon 2001). A translation of 1.8 mm along Z-axis and rotations of 1:3Æ and 0:5Æ around respectively X- and Y-axes have been detected. 1000 image pairs were recorded on both video-cameras giving after 3D reconstruction 1000 vector maps of 37 27 nodes.These instantaneous velocity vectors were averaged and rendered non-dimensional in order to 10 calculate the symmetrical self-similar mean proles hU i=U0 , hV i=U0 and hW i=U0 where U0 is the velocity along the jet axis. In the cylindrical coordinate system centered on the jet axis, the theoretical axial prole is given by 1 hU i = p (14) U0 (1 + S2 1 2 )2 where S is the spreading rate, = Y=(X X0 ) the non-dimensional radius and X0 the virtual origin (Pope 2000). SPIV using telecentric lenses 2 0.05 <V>/Uo 0.04 0.03 0.02 0.01 0 -0.01 0 0.02 0.04 0.06 0.08 η 0.1 0.12 0.14 0 0.02 0.04 0.06 0.08 η 0.1 0.12 0.14 0.05 <W>/Uo 0.04 0.03 0.02 0.01 0 -0.01 Figure 7. Non-dimensional mean out-of-axis velocities hV i=U0 , hW i=U0 of the round free jet v.s. the non-dimensional radius (Reynolds number of 21000, X=d in the 28-35 range) : ts of all the measured proles (continuous line) and theoretical proles (dashed line) The velocity U0 and the spreading rate S = 0:084 were extracted from Wygnansky and Fiedler (1969). The light sheet assumed to be vertical was assessed to be 5 mm behind the jet axis. The radius separating the measurement nodes from the jet axis is in the range from 0.033 in the middle of the view eld to 0.113 at the borders. Figure 6 shows that the measured and theoretical mean U-proles well overlap. The non-dimensional RMS deviation of the U-component is 0.0066. The measured V- and W-proles reported on gure 7 are in good agreement with the theoretical ones. The non-dimensional RMS deviations of the V- and W-components are respectively 0.0018 and 0.0035. One can notice that the V-component is null and the W-component maximal 11 at the minimal value of . When the radius exceeds 0.08, both V- and W-components decrease to zero. SPIV using telecentric lenses 5. Conclusion Telecentric lenses which form images by parallel projection allow to overcome diÆculties lied to the use of entocentric (conventional) lenses in stereoscopic PIV. In comparison to the classical SPIV methods, the (angular) telecentric method can support a high viewing angle contrary to the translation method and does not present a systematic image deformation as the angular displacement method does. In addition, when imaging through a liquid-air interface at right angle, if necessary by using a liquid prism (Prasad and Jensen 1995), the smear in the image plane produced by the light scattered from a particle remains circular away from the lens axes. In return, the limitation of telecentric lenses is the maximum size of the eld-of-view which is restricted by the diameter of the lens entry. In practice most telecentric lenses cover less than 15 cm. From the two telecentric congurations : the object-sided and the telescopic congurations, only the telescopic one forms images by parallel projection in Scheimpug condition. Combining the aÆne video-camera model with the 2D-3D mapping approach makes appear distortion compensation as an additive correction of both mesh points and displacement eld. The ability of this combination has been demonstrated on translation tests and on measurements in a free jet. When the correction of the displacement eld is negligible, such a method leads to a linear reconstruction using the constant magnication matrices. SPIV measurements can so be achieved within 0.1 pixel at standard magnication on a 1/2" video-sensor as the telecentricity is about 0:1Æ or less. In this frame, the technique can be successfully implemented for in-line controls of ows. References Adrian R. J. 1991 Particle-imaging techniques for experimental uid mechanics Annual Review of Fluid Mechanics 3 261-304 Coudert S. and Schon J. P. 2001 Back projection algorithm with misalignment corrections for 2D3C stereoscopic PIV Measurement Science and Technology 12 1371-1381 Fournel T., Coudert S., and Riou L. 2000 Stereoscopic 2D3C DPIV with telecentric lenses : calibration and rst results Proc. Euromech 411 Hinsch K. D.,Hinrichs H.,Roshop A.,Dreesen F. 1993 Holographic and steroscopic advances in 3-D PIV. Holographic Particle Image Velocimetry Proc. Fluids Engineering Division, American Society of Mechanical Engineers, ed. E.P. Rood 148 33-36 Hinsch K. D. 1995 Three-dimensional particle velocimetryMeasurement Science and Technology 6 742753 Konrath R. and Schroder W. 2000 Stereoscopic Particle-Image Velocimetry : A new approach using telecentric lenses Proc. of 10th International Symposium on Applications of Laser Techniques to Fluid Mechanics Konrath R. and Schroder W. 2002 Telecentric lenses for imaging in particle image velocimetry: a new stereoscopic approach Experiments in Fluids 33 703-708 Lawson N. and Wu J. 1997 Three-dimensional particle image velocimetry: error analysis of stereoscopic techniques Measurement Science and Technology, 8 1455-1464 SPIV using telecentric lenses 12 Lecerf A., Renou B., Allano D., Boukhalfa A. and Trinite M. 1999 Stereoscopic PIV : validation and application to an isotropic turbulent ow Experiments in Fluids 26 107-115 Pope S. B 2000 Turbulent ows Cambridge University Press 96-134 Prasad A. K. 2000 Stereoscopic particle image velocimetry Experiments in Fluids 29 103-116 Prasad A. K. and Jensen K. 1995 Scheimpug stereocamera for particle image velocimetry in liquid ows Applied Optics, 34 7092-7099 Solo S., Adrian R. and Liu Z.C. 1997 Distortion compensation for generalised stereoscopic particle image velocimetry Measurement Science and Technology 8 1441-1454 Tsai R. Y. 1987 A versatile camera calibration technique for high-accuracy 3D machine vision metrology using o-the-shelf TV cameras and lenses IEEE Journal of Robotics and Automation RA-3(4) 323-344 Van Oord J. 1997 The design of a stereoscopic DPIV system Report MEAH-161 Delft the Netherlands: Delft University of Technology Westerweel J., Dabiri D. and Gharib M. 1997 The eect of a discrete window oset on the accuracy of cross-correlation analysis of a digital PIV recordings Experiment in uids 23 20-28 Willert C. 1997 Stereoscopic digital particle image velocimetry for application in wind tunnel ows Measurement Science and Technology 8 1465-1479. Wygnanski I. and Fiedler H. 1969 Some measurements in the self-preserving jet Journal of Fluid Mechanics 38 577-612 Acknowledgements This work has been performed under the EUROPIV 2 project. EUROPIV 2 (A Joint Program to Improve PIV Performance for Industry and Research) is a collaboration between LML URA CNRS 1441, DASSAULT AVIATION, DASA, ITAP, CIRA, DLR, ISL, NLR, ONERA, DNW and the universities of Delft, Madrid, Oldenburg, Rome, Rouen (CORIA URA CNRS 230), St Etienne (TSI UMR CNRS 5516), Zaragoza. The project is managed by LML URA CNRS 1441 and is funded by the CEC under the IMT initiative (CONTRACT N: GRD1-1999-10835). The authors thank Pr K. Hinsch and Dr H. Royer for their helpful remarks. SPIV using telecentric lenses 13 Figure captions Telescopic system. Principal rays (dashed line) are parallel to the optical axis on both object and image sides. Figure 1. System telecentric on object side only. Such a system is more compact than the telescopic one. Principal rays (dashed lines) are no longer parallel to the optical axis on image side. Figure 2. Figure 3. Scheimpug condition for telecentric lenses Figure 4. Distortions of telecentric lenses RMS deviations X (triangles), Y (squares) and Z (diamonds) for out-of-plane displacements of a particle-seeded resin lighted with a laser sheet Figure 5. Non-dimensional mean axial velocity hU i=U0 of the round free jet v.s. the non-dimensional radius (Reynolds number of 21000, X=d in the 28-35 range) : t of all the measured proles (continuous line) and theoretical prole (dashed line) Figure 6. Non-dimensional mean out-of-axis velocities hV i=U0 , hW i=U0 of the round free jet v.s. the non-dimensional radius (Reynolds number of 21000, X=d in the 28-35 range) : ts of all the measured proles (continuous line) and theoretical proles (dashed line) Figure 7.