Experiments in Fluids 18 0995) 421-428 9 Springer-Verlag1995

On the accuracy of velocity and vorticity measurements with PIV
L. Lourenco, A. Krothapalli


Abstract A number of numerical techniques aimed at
improving the accuracy of measurements using the correlation
approach in Particle Image Velocimetry, PIV, are proposed
and investigated. In this approach the velocity (displacement)
is found as the location of a peak in the correlation map. Based
on an experimental model the best performing peak finding
approaches are selected among different strategies. Second, an
algorithm is proposed which minimizes errors on the estimates
of vorticity using velocity distributions obtained by means of
PIV. The proposed methods are experimentally validated
against a flow with known properties.

The measured velocity components are u* = u + e and
v*=v+e, where e denotes the absolute error. One way to
obtain the out-of-plane component of the vorticity field
consists of evaluating the velocity derivatives with a suitable
finite differencing scheme, such as the second order scheme in
Eq. (1)

One of the challenges in experimental fluid dynamics is the
measurement of the vorticity field. This difficulty arises from
the fact that vorticity is a quantity defined in terms of the
velocity gradients. Hence, the velocity measurement must be
made simultaneously over several closely spaced locations
from which spatial derivatives can be evaluated using finite
difference schemes. Among the currently available velocity
measurement techniques, Particle Image Velocimetry (PIV) is
the most suited for this kind of measurement as it is capable of
providing the velocity vector over a selected two-dimensional
region of the flow, with sufficient accuracy and spatial
resolution. This constitutes a great asset for the study of
a variety of flows that evolve stochastically both in time and
space, such as unsteady separated, transitional and turbulent
flows. This article discusses the accuracy that can be achieved
in the measurement of velocity and vorticity using PIV.

where ~R stands for higher order terms.
Equation (1) shows that the error in the measurement of the
vorticity depends on two components of different nature. One
is the truncation error associated with the finite difference
scheme used, of order (Axa, Ay2), the other due to the
uncertainty, e, in the velocity measurement and of order
(llAx, llAy). Whereas the truncation error decreases for
smaller grids, the opposite happens to the error due to the
velocity uncertainty. This situation suggests that for a given
velocity measurement error, e, there is an optimum spacing
(grid) for which the total error is minimized. It also shows that
meaningful vorticity estimates can only be obtained if the error
in the velocity measurement, e, is kept small.
Uncertainties in the velocity measurement include errors
introduced during the recording of the multiple exposure
photograph, such as the ones introduced by the distortion of
the scene being recorded by the camera lens, limited lens and
film resolution, three-dimensional effects (Meynart and
Lourenco 1984), bias introduced by large velocity gradients
(Adrian 1988, Keane and Adrian 1989), and the inaccuracies
due to the processing algorithms. In this article, schemes
aimed at the minimization of the error contribution due to the
processing algorithm are discussed. In addition, an adaptive
mesh scheme for the evaluation of the velocity derivatives is
proposed. This scheme is designed to minimize effect of the
local velocity measurement errors on the velocity derivatives
and hence the vorticity estimate.


Nature of the problem
Let us consider that the in-plane velocity components, u(u, v),
of a two-dimensional region of a given flow field are measured
with PIV, in a Cartesian grid, at uniformly spaced intervals.
Received: 16 May ~9941Accepted: 4 December 1994
L. Lourenco, A. Krothapalli
Fluid Mechanics Research Laboratory,
FAMU/FSU College of Engineering,
P.O. Box 2175,
Tallahassee, FL 32316-2175, USA
Correspondence to: L. Lourenco
Work supported by NASA Ames Research Center

\(u~+~-- u:- +-A)g ~'R+ A-y


The PlY technique
The measuring principle of Particle Imaging Velocimetry
consists of obtaining the velocity of a fluid by measuring the
velocity of tracer particles in suspension. This is accomplished
by illuminating a planar region of the flow with high intensity
strobe illumination, usually a pulsed laser, and recording the

(Dr) 2 {1 + cos [2~(Axoox+Ayc@)] } (3) coxand coyare the spatial frequencies along the x andy axis. (3). Kodak 2415 Technical Pan. For improved accuracy. 6(x.5 and F # of 2. such as Particle Tracking. the calibrated displacements were introduced via precision translation and rotation stages (Klinger Scientific models MRL8. i.y) of the particle image pairs.. y)| [5 (x. In this two-dimensional array. In order to minimize systematic errors introduced by such effects as film shrinkage. with fine grain. The "calibration" displacement fields are obtained by recording speckle patterns undergoing plane.coy) are the Fourier transforms of I(x. is designated as the data peak. its derivatives are equal to zero. for plane field recording. To ensure a high degree of repeatability. coy)12= 2 I*(09.y) can be represented as the sum of several copies of the function H(x. 4 Peak finding algorithms The displacement ~(A x. Equation (4) shows that the autocorrelation A (x. The refined peak position is taken where the fitting function has maximum. y + Ay) + H(x-.2~zj(xcox+ycoy) ] dcoxdcoy is the Fourier transform of the intensity halo I-/'~'[2. is obtained by taking the Fourier transform of Eq.e. b. However. A parabolic and a Gaussian fit are employed. whereas H(x. The relative accuracy given by these different approaches were investigated using "calibration" displacements. The film resolution is 320 line pairs/mm and is sufficient to resolve. i.5 degrees rotation and 100 pm translation. designated as H(x + Ax. There are several methods available to convert the data contained in the photographs. The spectrum. We name the function 2 H(x. excluding the self correlation peak. The Whittaker's signal reconstruction technique is used. and I* (co. Meynart and Lourenco 1984. The coefficients of the fitting function are found by least squares. The peak position is approximated by the centroid of the autocorrelation function in the vicinity of the data peak. digital processors or a combination thereof. In this approach the two dimensional autocorrelation of small interrogation regions is obtained.tracer position as a function of time in doubly or multiple exposure photographs. o&) 12exp [ .8.. What differentiates PIV from other whole field velocity measurement techniques.y) and D(x.y) centered at the origin of the correlation map. this estimate needs to be further refined.into velocity data (Meynart 1983. the average image intensity distribution. The common feature of any of these :. y) + 6 (x + Ax.e.Ay) (4) where j = x / ~ and H(x. This lens was chosen because of its optimal performance.y) represents the recorded images of the particles. of the particles recorded within the measuring volume. c.y) is the Dirac delta function centered at (x. Assuming two dimensional motion. is that in ideal operating conditions the tracer concentration is high enough such that a large number of particles exist within a specified measuring volume (Lourenco and Krothapalli 1987).. Ay) is given by the peak position of the continuous autocorrelation function A(x. y+ Ay). The lens was set at a magnification of 0. excluding the self-correlation peak. corresponding to 0. or alternatively H(x--Ax.y--Ay). Because the autocorrelation is represented digitally. A (x. y) = I(x.e. The primary objective of this study is to evaluate different strategies that can be used to accurately determine the position of the correlation peak with sub-grid resolution.e. peak. a~)12exp [ -. is proportional to the local displacement. and carried out using a series of "calibration experiments". the ensemble of particle image doublets is well represented (Lourenco et al. where the flow is considered to be sufficiently uniform.01 degrees respectively.Ax. The digital autocorrelation around the peak location is computed in a refined grid using an interpolation scheme. The displacement fields are recorded onto photographic film. by means of a 35 mm Nikon F-3 camera fitted with a 105 mm Macro lens. uniform rotation and translation. which provided accurate displacements within 0. + Ax.25 and TR80). Finally. The measuring volume is a small region within the fluid. which is well approximated by a Gaussian function (Goodman. minimum distortion. To further reduce the effect of image aberrations only the central 1/2 of the recording was used for the processing. 1968). coy)and D*(cox.5 ~tm. The "local" velocity is taken as the average velocity. i. Adrian and Yao 1984). y + Ay). The speckle patterns were generated by doubly exposing an aluminum surface.y).C. or group velocity. ID*(ro. The autocorrelation function is efficiently evaluated using two 2D Fourier transformations.y) = I~ [/*(co.y) and | is the convolution operator. y + Ay) ] (2) where I(x. to the coherent illumination of a HeNe laser. (2) and multiplying with its complex conjugate. i.processors is that the autocorrelation is obtained in a digital array format. we opted to compare how well the different algorithms track the change of displacement with . y .21rj (xcG +ycor) ] dcoxdoay = 2H(x. the maximum displacement is initially obtained by the location in the discrete correlation array with the maximum value. y) + H(x + Ax.y). the visual Young's fringes. The velocity is given by the ratio between corresponding tracer images and the time between strobe pulses. the D. To perform these transforms different schemes are used that employ optical processors. Figure 1 is an example of one of such recordings.. the most commonly used techniques use the autocorrelation approach. i. the autocorrelation is obtained as the Fourier transform of Eq. 1994) by D(x. Three different strategies are investigated: a. This is achieved using an experimental approach. in order to resolve the peak coordinates with subgrid accuracy.e.1 ~tm and 0.y) = jj ID*(09x. alias free diffraction limited speckle images with average size of 6. the location of the maximum value in the array. The "average" displacement within the interrogation region corresponds to the coordinates where the autocorrelation function has a maximum value. The digital autocorrelation in the vicinity of the peak is approximated by a known continuous function.

H(x + Ax. 1. centered at the element with maximum value. y + dy) is clearly separated from H(x.1 Centroid method Equation (5) represents the average displacement computed using the centroid method. times the magnitude of the imposed rotation. Thus the centroid method should only be used in situations when the displacement is such that the function H(x + dx.423 Fig. H(x. good agreement is obtained between measured and actual displacement data. y + Ay) M ~ H(x + Ax. Ay) = ~ ~ (x.y + dy) changes. This position is found. This lack of performance is not surprising and is accounted for by the fact that the shape of the correlation peak. Figure 2 is a typical result showing the relationship between the actual displacements. In this study the autocorrelation was obtained using a combination of optical and digital processors. pedestal. and a second Fourier transform is computed digitally to yield the autocorrelation.C. and the correlation peak are well approximated by a Gaussian. herein defined as the D. y) H(x + Ax. as shown by the ladder like behaviour in Fig. In a preceding paragraph it has been shown that the image intensity distribution. Figure 3 shows that this interpolator can produce large errors as it tends to bias the results towards displacements that correspond to discrete integer frequencies. The Gaussian fit assumes that the correlation function in the vicinity of the data peak is . and allowed us to measure small displacements in the vicinity of center of rotation. For small displacements. peak.C. 4. 2. with sub-grid accuracy. 3. when a parabolic fit in the least squares sense is used. and set where the sign of the derivative of H(x + Ax. The discrepancies observed in the low range of displacements can be accounted for by the contamination of the data peak by the D. y). The limits of integration are variable. The Young's fringe pattern is digitized in the 256 x 256 x 8 bit pixel format. using the various methods which are described in the following. The translation was primarily used as a means of increasing the dynamic range of the technique. peak. making 250 -~ 200 ~ 150 ~~. r . Thus the displacements were measured along an axis with the origin at the center of rotation. is poorly approximated by a parabolic shape. defined as (r-ro)'r~. the displacement which has ben recorded in the region illuminated by the laser beam is proportional to the location of highest amplitude peak of the digital autocorrelation excluding that at the origin. Accordingly. 4.C. This result shows that for displacements larger than a limit value. In this approach a collimated laser beam with 300 ~tm in diameter illuminates the film negatives to produce the interference Young's fringe pattern in the focal plane of a converging lens. Measured Displacements (Centroid Scheme) the D. y + Ay). and the measured displacement obtained as the location in the autocorrelation map. g( Ax. r~. Calibration displacement field position as imposed by the rotation. to.y) to become superposed to the data peak. Actual vs.2 Curve fitting Figure 3 shows the actual displacement versus the measured displacement.r0.100 ~. Ax~O and dy~O. 50 0 S f 100 150 200 Actual disptocemen{ (~m) 250 Fig. In the data that is presented in this article this "bias" effect is removed after processing. and compared with actual displacement values given by the distance from the axis. This interpolator uses a 3 x 3 computational cell.y+ Ay) (5) where M is the magnification factor.

As a result the autocorrelation is obtained with half the spacing. . 100 150 200 Actual d}splecement (pro) 250 . -. or if excessively long illumination times generate streaks. For example if the fringe pattern is originally obtained in a 128 x 128 format. 4. f 150 f ' i . 6. Measured Displacements (Gaussian Fit with Padding) closely approximated by H(x.y)=Cexp[ (x-dx)2 (y-Ay)2q ~2 ~ j (6) where C and ~ are the fitting constants. 4. 200 'E ~ 150 150 N r~ .y) An alternative way of increasing the resolution of the correlation axes is padding the digitized fringe pattern with zeros (Searns and Hush. thus confirming the previous analysis. Actual vs. . Measured Displacements (Parabolic Fit with Padding) Fig. Actual vs. In contrast to the previous results this figure shows excellent agreement between actual and measured data.~_ 100 ~ 100 "~ 5o ~ 50 (/1 a I 0 0 100 150 200 Actual displacement (izm) I 0 250 Pig. . The reason for improved accuracy with the Parabolic fit/Padding . this approach is costly in time because it increases the length of the second 2D FFT operation by a factor of four. -~ 200 =1. However it is expected that the accuracy provided by this interpolator is decreased when the intensity distribution of the recorded image is not approximated by a Gaussian shape. (6). I 100 150 200 Actual displacement (/~m] 50 250 E . as it requires very few computations.250 250 -~200 =t. . then it is appended with zeros to become 256 x 256 in size. 5 and 6). The main advantages of the Gaussian interpolator are its efficiency. 4.3 Padding the FFT'S ~(x. Actual vs. This can be the case when the images produced by the tracers are affected by distortions due to lens system. Actual vs. However. E 150 ul 100 ~100 o 501 ~ 5O I 0 P 50 .200 E =1. 5. . Changing the fitting function to a Gaussian yields the results in Fig. 3. Use of padding with either the Parabolic or the Gaussian fit showed remarkable improvements in accuracy (Figs. Measured Displacements (Parabolic Fit) 250. and accuracy. ' ' i . Measured Displacements (Gaussian Fit) I I 100 150 200 Actual disptacement (Fm) 250 Pig. 1990). The Gaussian fit is reduced to the parabolic fit problem by taking the logarithm (linearizing) of Eq. 250 Fig.

Detailed comparisons show that this Original.m) 250 Fig. grid First sub-grid 2 i. with one fourth the original spacing. H(k2. To further evaluate its accuracy under real measurement conditions a control experiment was carried out. laminar wall-jet issuing from a rectangular . The amplitude of the autocorrelation at any subpixel location is given by two-dimensional reconstruction as: i+2 ~ 150 8 100 425 g ~ 50 /+2 H(x. ~. considered identical. at eight intermediate locations (1 through 8). position scheme retains the same accuracy obtained with the Gaussian fit/Padding approach but with considerable savings in computation time. e 9 9 . The autocorrelation is again recomputed in a refined grid.12y) k=i-2 l=j-2 I o / 50 I 100 i I 150 200 250 Actuot disptocement (Fm) ~(x-k2x) ~(y-. A new maximum coordinate position is selected among the previous maximum and these new data points.u. As shown in Eq. In this series of experiments the errors are bounded and of the order of _ 1 pm." t" 9 ~o o 9 "o 9 9 _- . Actual vs. 9.4 ~. The value of the autocorrelation is evaluated in a new grid.--o' 9149 9149 #o 9 9 -to ~ 9 ~ 1 7 6 1 7 96 -o 9 oO r - 9 9 9 9 9 9 9 "t i P 5O I I I 100 150 200 Actuo[ disp[ocement (. 4 5 Second sub-grid Fig.~y) where 2~ and 2y are the sampling intervals. in the x and y directions respectively.. Measured Displacements (Whittaker's interpolation) 3 9 2 : . The approximate location of the x and y components of the displacement is first given by the discrete coordinates of the maximum intensity of the autocorrelation (i. 7.combination is due to the increased closeness of the discrete data to the true maximum of the continuous autocorrelation. The signal reconstruction technique known as Whittaker's reconstruction or Cardinal interpolation is the scheme chosen in this investigation (Stearns and Hush.j 6 3 Fig. The experiment consisted of the measurement of the velocity and vorticity distribution of a two-dimensional.y)= ~. Figure 9 shows the measurement error as a function of the position along the interrogating axis. j 1 ~ 0 t~ -1 ".. j). Grid for the Whittaker's interpolator 5 Experimental verification: control e x p e r i m e n t The Whittaker's reconstruction peak search algorithm was selected on the basis of its improved performance versus the other techniques. This procedure is repeated until the desired resolution is achieved. i and j are indices corresponding to the array element with the maximum value. followed by a reselection of a new peak position. Figure 8 shows a typical result when the Whittaker interpolator is used. incompressible. 200 Whittaker's reconstruction Because the Zero Padding technique is computationally demanding it is desirable to employ alternative interpolating schemes that offer comparable accuracy while remaining computationally efficient. 8. 250 4. with half the spacing. Error vs. 7. (7) the summation is carried over a localized 5 x 5 cell region. The application of this scheme is illustrated in Fig. Typically. six of these cycles are performed to achieve a nominal accuracy of 1/64 th of a pixel. 1990).

between the measured velocity distribution. and the abscissa with the location corresponding to the maximum velocity. Figure 12 presents a comparison between the vorticity distribution evaluated by means of the second order scheme and that computed using the numerical simulation. provides the illumination sheet. Although the general trend of the profile is found. Wall-jet velocity profile at x/h = 3 simulation at a downstream location corresponding to three widths. Additionally. 6 Computation of vorticity: adaptive scheme The normalized vorticity distribution corresponding to the velocity profile in Fig. Therefore it is expected that a smaller spacing will produce better estimates of the vorticity distribution. Doubly-exposed photograph of plane containing the jet's mid axis channel into still air. Measurements are carried out within twenty widths from the exit. 11 was first evaluated using the second order accurate central-difference scheme. 11. interface between the primary and entrained fluid depicted in the photograph clearly demonstrates that. frequency doubled Nd-Yag laser (Lumonics HY-400). 1986). using conventional photographic PIV and the numerical ] 1. it is clear that individual estimates of the vorticity can be affected by large error. A velocity bias device. has a parabolic shape. covering the region from the jet exit up to 3. 10.8 O. Similar levels of agreement were obtained. 1987) and set at 18. heavily seeded and the lightly seeded ambient still air. A single double-pulsed. The clear. 1-5 ~tm in diameter. The flow Reynolds number based on the jet width and the average mass flow velocity is 1. In this scheme the truncation error has a contribution of order (Ax 2. is used to accommodate the large velocity range within the flOW. For the PIV measurement. The time separation between the double light pulses is adjusted according to the maximum flow velocity (Lourenco and Krothapalli. the free-shear layer and the boundary layer. a numerical simulation of the ftow using the Boundary Layer approximations is available.2 0 0 I P I 2 3 Y/Y~• Fig. showing good agreement. This photograph clearly shows the two fluid streams that form the jet: the primary jet.3 widths downstream. This comparison was carried at downstream locations where the flow remained two-dimensional.9 ~tseconds. As it will be discussed in the next section. the primary air jet and the ambient air are seeded with small oil smoke particles. and a wide velocity dynamic range. In this experiment the jet width is 5 mm.4 0. in the near wall-region.0 o - PIV Numeric. the amplitude of flow instabilities can be neglected. at the end of the channel.Fig. This simplistic analysis does not account for the contribution of error affecting the velocity measurement discussed in the introduction. The particle image doublets are also visible. This test flow is chosen because it contains two regions with large velocity gradients. riple free. . Figure 10 is a typical doubly exposed photograph of the central plane of wall-jet. has an aspect ratio of twenty and the wall is one hundred widths long. where the flow can be considered two-dimensional. the real challenge remains the evaluation of the vorticity from the velocity data with minimal error. Ay2). in the initial region of jet. up to a maximum near the jet axis. For these conditions the exit velocity profile. Thus comparison with the numerical solution for laminar flow is appropriate. consisting of a scanning mirror (Lourenco et al. This numerical solution is second order accurate in both the x (streamwise) and y (cross-stream) directions and is used as a means to test and validate the accuracy of the PIV results.ol 0.300.6 0. Results from the numerical code compare extremely well with Glauert's similarity solution for fully developed flow. The image displacement on these frames ranges from a minimum (corresponding to the velocity bias magnitude). In this figure the velocity is normalized with the maximum velocity. The doubly exposed frames are captured by means of a Nikon 35 m m camera equipped with a 50 m m macro lens. at a time instant. from zero near the wall. Figure 11 is a typical comparison. to a maximum near the jet axis.

. j~-)~*. is smaller than the difference between estimates with the same truncation error. Therefore application of the scheme stops when the error due to the experimental uncertainty exceeds the truncation error.2 J This scheme is applied in the same manner to generate another approximation of the derivative with fourth order truncation error using grid spacing 2h and 4h.g. ell. and thus the accuracy of a derivative estimate. using the adaptive scheme and the numerical simulation is obtained. Further increase in accuracy in the evaluation of the derivative can be achieved if the derivative estimates j~. by combining values of derivatives evaluated in two different grids as follows. )~*-f2*.096al h 4 q.. for example the differences formed with 3~-f4 and ~-f4*. 12. In this case the new derivative estimate becoming j~** and so on. The grid is then further refined until the contribution of the experimental error becomes larger than that of the truncation error. Figure 14 presents the vorticity distribution when this approach is implemented.g. (12) h2 h4 8h 4h )~ A* 2h h ]~ f j~* f* h6 f2** f ** ha f *** Error due to experimental uncertainty e/8h e/4h e/2h e/h . Wall-jet vorticity distribution (central-difference scheme) A differencing scheme aimed at the minimization of the total error affecting the vorticity estimate is proposed. 3 7 6 a 2 n + . For example the experimental uncertainty affecting the f and f * estimates are the same. Richardson's extrapolation fexact"~~ "~ 16a0h 2 + 256alh4+ -. In practice this is accomplished by comparing the difference between estimates with the same order truncation error and those with same experimental error. 3~ and )~ are computed using the least squares second order polynomial approximation to the data set {u~}.~ + 4.-6 = ]-~ + 7 ~ + 5 . i + 4 and i . This scheme is based on the Richardson's extrapolation principle. The grid is further refined as long as the difference between estimates affected by the same experimental error. In this scheme the derivatives of the velocity are first computed in coarse grids.8 . However it is important to note that the error due to the uncertainty in the measurement of the velocity remains approximately constant for approximations of the derivative involving the same grid spacing. e. and f4 the divided central difference obtained for a spacing equal to 4h: fa=uJ+8--uJ-a-fexact + .=~ (8) eV = aoh 2 + ai h4 ~. (14) -I I 2 Y/Ymo~ Fig.. ~=uj+4-uj-4 8h (10) Table 1..a2 h6 + "'" (9) The application of Richardson's extrapolation increases the order of the truncation error. This estimate is denoted as ~*: t 9 o~ -~x~ ~ 9 PIV f2* 4~-f4 7e --3 q-12-h +64alh4+ "'" (13) Next the estimates f4* and J~* are used to eliminate the fourth order truncation error to yield: 0 427 f2** 16f2*--f4* 49e . in order to keep the error contribution due to the experimental error. relatively small when compared to the truncation error er. as more data points are considered to approximate the function. 13. In this fashion the contribution of the error e is diminished.. The experimental and the truncation errors are expressed as functions of h..4 and i + 2 and i ... Consider that • is the divided central difference of the function u (velocity component) obtained with a large data spacing equal to 8h.. (11) where fexac. This scheme can be applied in succession to eliminate the higher order truncation errors as shown in Table 1. Very good agreement between the vorticity distribution obtained from the differentiation of the PIV data. The second order term can be eliminated from the two equations to yield a better estimate of the derivative as: f4* 4f4--fs 7s = 3 +2--~ + 1. and shown in Fig.2 respectively. the grid spacing: 8. denotes the exact value for the derivative. e. In this manner the local velocity is first represented by a polynomial which includes information from all the points contained between i + 8 and i ..024alh4 + ..

with data obtained by means of PIV. Conclusions o " -- -o~ A detailed study on the accuracy of different peak interpolating algorithms was carried out.g.231-239 Meynart R. Orlando: Laser Inst Amer 141-161 Lourenco L. Belgium Meynart R (1983) Speckle velocimetry study of vortex pairing in a low RE unexcited jet. The scheme can be applied to other methods of computing the vorticity. LIA Proc. Vol. Appl Opt 33:2465-2470 Lourenco L. Buchlin JM (1986) A non-invasive experimental technique for the measurement of unsteady velocity and vorticity fields. New York: McGraw-Hill. Exp Fluids 5:29-32 Lourenco L. The Gaussian peak interpolator and the Whittaker's interpolator. Missouri. 115-129 Adrian R]. PIV Numerical simulation O 0 "1 i i i 1 2 Y/Ymax Fig. Yao CS (1984) Development of pulsed laser velocimetry (PLV) for measurement of turbulent flow. Riethmuller ML. and hence the vorticity. as long as they produce vorticity estimates affected by truncation errors that are expressed by Eq. Second edition. Gogineni SP. in a laminar wall-jet. 68. 170-186. It is demonstrated that this scheme provides better estimates of the vorticity distribution for the wall-jet. Krothapalli A. Vol.. Based on the findings of this study two approaches were recommended. Adrian RJ (1989) Optimization of particle image velocimeters. Goodman ]W (1968) Introduction to Fourier Optics. III. In: Optical methods in flow and particle diagnostics. 14. AGARD-CP-413. von Karman Institute Lecture Series. The performance of these approaches was further verified by comparing a well established flow field velocity distributions. 75-84 . was proposed. Lourenco L (1984) Laser Speckle Velocimetry.. 287 Keane RD. Springer Verlag. Prentice Hall. (9). e. the circulation approach. j . In: Proc Symp Turbl Rolla: Univ. Phys Fluids 26:2074-2079 Stearns SD. In addition a new scheme for the computation of velocity derivatives. j 7 . LaSalle RT (1994) On-line particle image velocimeter: an integrated approach. Hush D (1990) Digital Signal Analysis. Wall-jet vorticity distribution (adaptive scheme and least squares) Adrian R] (1988) Statistical properties of particle image velocimetry measurements in turbulent flow. Brussels. 13. The selection of these algorithms was based on the analysis of the signal being interpolated as well as on some experimental models. In: Laser Anemometry in Fluid Mechanics. Wall-jet vorticity distribution (adaptive scheme) References - o - PIV Numerical simulation 0 -1 0 1 2 3 YIYmax Fig. Krothapalli A (1987) The role of photographic parameters in laser speckle or particle image displacement velocimetry.