Professional Documents
Culture Documents
This article has been downloaded from IOPscience. Please scroll down to see the full text article.
(http://iopscience.iop.org/0957-0233/8/8/010)
View the table of contents for this issue, or go to the journal homepage for more
Download details:
IP Address: 81.31.173.171
The article was downloaded on 30/10/2010 at 10:22
Received 8 January 1997, in final form 23 April 1997, accepted for publication
2 May 1997
Abstract. A geometric error model for analysis and design of stereoscopic PIV
systems is presented. The model allows displacement errors in either translational
or angular systems to be analysed for any given angle or camera separation and
for any off-axis position. A parameter for the analysis of the system performance is
also introduced based on the ratio of out-of-plane to in-plane errors. This is
subsequently used to investigate the relative performance of translational and
angular PIV systems for camera angles up to 45◦ and camera separations of half
the object distance. Results from this analysis show similar trends in centreline
characteristics for both types of stereo systems but different trends in off-axis error
ratios due to imaging geometry. The results have also suggested that a CCD-based
angular PIV stereo system offers up to 40% greater out-of-plane accuracy for a
given field of view and laser power than previous translational systems.
0957-0233/97/080894+07$19.50
c 1997 IOP Publishing Ltd
Error analysis of stereoscopic systems using 3D PIV
With reference to a previous analysis [17], let us consider Differentiating (1) and (2) in terms of x, y and z leads
a general case where two cameras, j = 1, 2, are imaging to particle image displacements (1Xj , 1Yj ) in the image
a common object plane in a fluid with coordinates (x, y, z) plane equal to:
and origin O and which have image planes (Xj , Yj ) as ∂Xj ∂Xj ∂Xj
shown in figure 1. Also let the two cameras be separated 1Xj = 1x + 1y + 1z (7)
∂x ∂y ∂z
by a distance hj along the x axes with respect to the lens
centres Oj and let the camera optical axes be rotated by ∂Yj ∂Yj ∂Yj
an angle αj with respect to the z axes. If both cameras 1Yj = 1x + 1y + 1z. (8)
∂x ∂y ∂z
have object and image distances do and di respectively,
To extract velocity data from the particle image
then for a given particle moving in the object plane, the
displacements, (1Xj , 1Yj ), the particle displacements in
corresponding coordinates in the image plane will equal:
the object plane (1x, 1y, 1z) must be found. Therefore if
Xj = −Mj xj0 (1) α1 = −α, α2 = +α, h1 = −h and h2 = +h then (7) and
(8) can be written in terms of x, y and z such that:
Yj = −Mj yj0 (2)
1X1 = a1 1x + b1 1z (9)
where the magnification Mj is given by
y
Mj = di /(do − zj0 ) (3) 1Y1 = −M1 1y + (1x sin α + 1z cos α) (10)
do1
and xj0 , yj0 and zj0 are transformed coordinates in the image 1X2 = a2 1x + b2 1z (11)
planes as shown in figure 1 such that: y
1Y2 = −M2 1y − (1x sin α − 1z cos α) (12)
do2
xj0 = −z sin αj + (x − hj ) cos αj (4)
where
yj0 = y (5) do1 = do − z cos α − (x − h) sin α (13)
895
N J Lawson and J Wu
and di are negligible and that the errors in the image plane Therefore the error ratio is independent of y and z but
(1X1 , 1Y1 ) and (1X2 , 1Y2 ) are equal such that: will vary in x. At the centre of the measured field where
x = 0, y = 0, z = 0, the error ratio will simply equal:
δ(1Xj ) = δ(1Yj ) = δ(1X) (26)
1
er = . (36)
for j = 1, 2 where δ( ) represents the error in a given (h/d )
o x=0,y=0,z=0
variable. Therefore using standard error analysis the This relationship is plotted in figure 2 and can be used as
uncertainties in the object planes, 1x, 1y and 1z can be an initial design guide for a translational system.
written q The off-axis error ratio relationship is more complex
b12 + b22 and is shown in figure 3 for a range of (x/do ) and (h/do ).
δ(1x) = δ(1X) (27) These results show the error ratio significantly falling away
|a1 b2 − a2 b1 |
q from the centre of field. This reduction in er is caused by
δ(1y) = c12 + c22 + c32 δ(1X) (28) the effective increase in viewing angle for a corresponding
movement off-axis. The increase in viewing angle results
q in greater values of 1X1 and 1X2 in the image plane for a
a12 + a22 given particle movement, therefore improving the estimate
δ(1z) = δ(1X) (29)
|a2 b1 − a1 b2 | of 1z and hence er .
896
Error analysis of stereoscopic systems using 3D PIV
Figure 2. Translational system centreline error ratio versus camera position (x = 0, y = 0, z = 0).
Figure 3. Translational system off-axis error ratio for a range of h /do and y = 0, z = 0.
5. Angular set-up This result is more complex than the translational system
in described in equation (35) with a dependence on z as
If we now analyse the performance of an angular set-up also well as x and α. At the centre of the measured field where
with two identical cameras where h = 0, M1 = M2 = M x = 0, y = 0, z = 0, the error ratio will simply equal:
and do1 = do2 = do , then from equation (33) the error ratio
er can be written: 1
er = . (38)
tan α x=0,y=0,z=0
er = {cos2 α − sin2 α[2(z/do ) cos α + (z/do )2 sin2 α
+(x/do )2 cos2 α]}1/2 {sin α[1 + 2(z/do ) cos α This relationship, which is plotted in figure 4, has a similar
+(z/do )2 sin2 α + (x/do )2 cos2 α]1/2 }−1 |x,y,z . (37) trend to the to the translational characteristic shown in
897
N J Lawson and J Wu
Figure 4. Angular system centreline error ratio versus camera angle (x = 0, y = 0, z = 0).
figure 2 and can also be used as an initial design curve therefore giving a smaller improvement in the estimate of
for an angular system. 1z and hence er .
The off-axis error ratio relationship is shown in figure 5
for a range of (x/do ) and α. These results, as with the 6. Discussion
translational system, show er falling away from the centre
of field. The reduction in er , however, is not as significant The analysis in the previous section has defined error
as with a translational system. This is due to the rotation characteristics which result from the camera geometry of a
of the lens axis in the angular system which will reduce the stereo PIV system. These relationships can be used in the
variation in effective viewing angle across a given field. initial design stages of either translational or angular stereo
This results in a smaller change in values of 1X1 and 1X2 systems. There are, however, other major factors including
across the image plane for a constant particle movement image aberrations, laser power and image correspondence
898
Error analysis of stereoscopic systems using 3D PIV
which affect the accuracy of the stereo PIV system. example, if an angular system is required with an equivalent
Therefore assuming the use of correlation techniques to performance of the system described by Prasad and Adrian
extract the particle image displacements, these problems [9], i.e. a centreline error ratio of er = 3.5, then this
will now be discussed in the following section with corresponds to a camera angle of α = 16◦ . With this angle
reference to the results from the error modelling. and if using similar 500 ×400 format cameras with f numbers
The error modelling has shown for either translational of f 16, the depth of field will not match the viewed
or angular systems that large camera spacings or angles field with normally aligned camera backs. If, however,
are required to attain acceptable levels of error ratio er . the camera back is tilted to the Scheimpflug condition as
For example, to achieve a centreline error ratio of er = 2 demonstrated by Prasad and Jensen [14] then focusing will
requires a camera separation of h/do = 0.5 (α = 0) or be improved. Aberrations still occur in the image but these
a camera angle of α = 26.6◦ (h = 0). Unfortunately can also be reduced by the use of a liquid prism [14]. More
these geometries will cause image aberrations such as importantly with the angular system, the angle at which the
coma, astigmatism and distortion due to optical properties extreme rays enter the lens does not significantly change
of the imaging system [18] and will ultimately limit the with increasing α. Therefore in this case it is possible to
maximum camera separation or angles and hence 1z operate the angular system at α > 16◦ without having to
accuracy. These effects can be reduced by using small increase the f number. Prasad and Jensen [14] suggested
apertures and suitable lenses such as planar lenses with long optimum f numbers of f 8. Hence using the same laser
focal lengths [9, 10, 18]. Aberrations are also unfortunately power, a higher out-of-plane accuracy will be possible with
accentuated when imaging through a liquid–air interface an angular system.
such as when taking PIV measurements from a liquid If CCD cameras without any image plane tilt and with
system. In this case, previous authors have used various an array size of 8 × 6 mm2 are used to image a similar
methods to improve the image quality which include image field of view as the 500 × 400 format cameras, magnifications
plane tilt [9, 14] and use of a novel liquid prism [14]. of around M = 1/16 are required. These magnifications
The requirement for small apertures to reduce correspond to an acceptable depth of field at f 8. At f 16 the
aberrations demands greater laser power than for a 2D increase in depth of field would allow the camera angle to
system with the same field of view. Illumination levels be increased to α = 26◦ . This corresponds to a centreline
will also need to be increased when measuring from a liquid error ratio of er = 2 and is an improvement in out-of-
system due to the increased light loss by reflection through plane accuracy of 40% when compared to the previous
the liquid–air interface [9, 14]. Previous authors report the translational system [9]. Therefore, although a CCD system
use of f numbers as high as f 16 with camera spacings of will not offer the spatial resolution of a 500 ×400 film system,
h/do = 0.28 [9] and h/do = 0.2 [10]. These configurations it can offer improved out-of-plane accuracy for the same
correspond to centreline error ratios of er = 3.5 and er = 5 laser power.
respectively. With greater laser power, higher f numbers
such as f 32 could be used to increase camera spacings.
It is unlikely, however, that a translational system will be 7. Conclusions
able to achieve error ratios of er < 2 due to aperture and
laser power limitations. This paper has introduced a geometric error model for
Image registration and calibration are also major analysis and design of stereoscopic PIV systems. The
considerations for stereo PIV camera systems. Image model allows displacement errors in either translational
registration ensures that the two measurements 1X1 and or angular systems to be analysed for any given angle or
1X2 correspond to the same region in the fluid flow. camera separation and for any off-axis position. Subsequent
Previous workers [9] have successfully used a Moiré investigations used the model to examine the relative
technique to align both transparencies before processing. performance of both types of stereo PIV systems based
An alternative method has used a single camera with on an error ratio parameter. This error ratio was defined as
two sets of mirrors to record both images on the same the ratio of the out-of-plane to the in-plane errors. Results
transparency [10]. This greatly simplifies PIV image from the investigation have shown that the centreline error
registration. Other techniques applied to digital PIV ratios follow similar trends but off-axis error ratios follow
systems involve the use of a geometric transform and different trends due to the imaging geometry. Comparisons
bicubic interpolation at the image processing stage [12]. with previous work have suggested an angular-based stereo
In addition, image calibration is required to correct for the PIV system would offer greater out-of-plane accuracy for a
varying magnification caused by distortion of the image. given field of view and laser power whether using a film-
This is generally done by either displacing a calibration or CCD-based recording system. Improvements in out-of-
image in the object plane to produce a double-exposed PIV plane accuracy in this case may be 40% greater than an
image [10] or placing a known grid in the object plane and equivalent translational system.
then applying geometric transforms or bicubic interpolation
routines to the grid positions [9, 12]. These calibration
maps are then referenced during the correlation stage to References
calculate particle displacements. [1] Adrian R J 1991 Particle-imaging techniques for
The angular system offers the possibility of achieving experimental fluid mechanics Ann. Rev. Fluid Mech. 23
higher performance than a translational system. For 261–304
899
N J Lawson and J Wu
[2] Pickering C J D and Halliwell N A 1985 Particle image [10] Arroyo M P and Greated C A 1991 Stereoscopic particle
velocimetry: a new field measurement technique Optical image velocimetry Meas. Sci. Technol. 2 1181–6
Measurement in Fluid Dynamics, Inst. Phys. Conf. Series [11] Grant I, Zhao Y, Tan Y and Stewart J N 1991 Three
vol 77 (Bristol: Adam Hilger) component flow mapping: experiences in stereoscopic
[3] Mao Z Q, Halliwell N A and Coupland J M 1993 Particle PIV and holographic velocimetry Laser Anemom. Adv.
image velocimetry: high-speed transparency scanning Appl. 1 365–71
and correlation-peak location in optical processing [12] Westerweel J and Nieuwstadt F T M 1991 Performance
systems Appl. Opt. 32 (26) 5089–91 tests on 3D velocity measurements with a two-camera
[4] Meinhart C D, Prasad A K and Adrian R J 1993 A parallel digital particle image velocimeter Laser Anemometry
digital processor system for particle image velocimetry Advances and Applications vol 1, ed A Dybbs and B
Meas. Sci. Technol. 4 619–26 Ghorashi (New York: ASME) pp 349–55
[5] Coupland J M and Halliwell N A 1992 Particle image [13] Grant I, Fu S, Pan X and Wang X 1995 The application of
velocimetry: three-dimensional fluid velocity an in-line, stereoscopic, PIV system to 3-component
measurements using holographic recording and optical velocity measurements Exp. Fluids 21 214–21
correlation Appl. Opt. 31 1005–7 [14] Prasad A K and Jensen K 1995 Scheimpflug stereocamera
[6] Barnhart D H, Adrian R J and Papen G C 1994 for particle image velocimetry in liquid flows Appl. Opt.
Phase-conjugate holographic system for high 34 7092–9
resolution particle image velocimetry Appl. Opt. 33 [15] Keane R D and Adrian R J 1990 Optimisation of particle
7159–70 image velocimeters. Part I: double pulsed systems Meas.
[7] Raffel M, Gharib M, Ronneberger O and Kompenhans J Sci. Technol. 1 1202–15
1995 Feasibility study of three-dimensional PIV by [16] Prasad A K, Adrian R J, Landreth C C and Offutt P W 1992
correlating images of particles within parallel light Effect of resolution on the speed and accuracy of particle
sheets Exp. Fluids 19 69–77 image velocimetry interrogation Exp. Fluids 13 105–16
[8] Brucher C 1996 3D PIV via spatial correlation in a [17] Diner D B and Fender D H 1993 Human Engineering in
color-coded light sheet Exp. Fluids 21 312–14 Stereoscopic Viewing Devices (New York: Plenum)
[9] Prasad A K and Adrian R J 1993 Stereoscopic particle pp 49–65
image velocimetry applied to liquid flows Exp. Fluids 15 [18] Longhurst R S 1973 Geometrical and Physical Optics 2nd
49–60 edn (Longhurst) pp 17–41
900