A. M. Fincham, G. R. Spedding
Department of Aerospace Engineering,
University of Southern California,
Los Angeles, CA 90089-1191, USA
Present address: Laboratoire des Ecoulements Geophysiques et
Industriels, Coriolis, B. P. 53 X, F-38041 Grenoble Cedex 9, France
Correspondence to: A. M. Fincham
We would like to thank Prof. Ron Blackwelder for initiating this
research, and for providing continued support over the years.
Funding from the Ofﬁce of Naval Research, through the Fluid
Mechanics program administered by Dr. L. P. Purtel, has made
this work possible.
Experiments in Fluids 23 (1997) 449—462 Springer-Verlag 1997
Low cost, high resolution DPIV for measurement of turbulent fluid flow
A. M. Fincham, G. R. Spedding
An optimized cross-correlation based Imaging
Velocimetry system is described and its performance is
evaluated in numerical and physical experiments. Given
a discrete image array pair, the ﬂow seeding and image
processing parameters are optimized to maximize displace-
ment accuracy, regardless of the computational cost; collec-
tively these techniques are known as Correlation Imaging
Velocimetry (CIV). Order of magnitude improvements over
standard DPIV methods can readily be obtained, allowing high
resolution measurements to be made with low cost standard
resolution cameras. Fundamental limits on the measurable
range of length, velocity and vorticity scales are identiﬁed, and
related to those encountered in homogeneous, 3D turbulence.
The current restrictions apply to all imaging velocimetry
measurements; some paths for future research that are likely to
be proﬁtable are identiﬁed, together with some that are not.
Extensive use of CIV in this and other laboratories has allowed
direct veriﬁcation of these optimization principals.
DPIV (Digital Particle Imaging Velocimetry) methods, pion-
eered by Utami et al. (1984, 1987, 1990, 1991) and Willert and
Gharib (1991) have found widespread acceptance and have
been applied to a variety of ﬂuid mechanics problems (e.g.
Grant 1994). Compared with alternative, laser-based, analog
interrogation systems, DPIV methods, relying as they do on
non-specialized digital electronics, are both economical and
efﬁcient. They are also well-positioned to proﬁt from predict-
able (commercially driven) advances in technology.
Although the limitations and accuracy of various Imaging
Velocimetry (IV) methods have been discussed in certain
instances (most notably by Adrian 1988, 1991; Prasad et al.
1992; Westerweel 1993; Cowen and Monismith 1996), a com-
prehensive and general treatment is lacking. Limits on
accuracy are often constrained by the ﬂow parameters themsel-
ves, indicating the need for a speciﬁc IV error analysis
If DPIV methods are to be successfully applied to turbulence
measurement problems, characterized by 3D motion with
broad ranges of velocity and length scales, then close attention
must be paid to achieving the best possible accuracy, given
a certain sampling resolution. In essence, the question is
addressed; given a ﬂow phenomenon to be measured, and
some speciﬁc hardware conﬁguration, what is the maximum
amount of useful information that can be extracted by image
There are three objectives in the paper:
(1) To present and specify a set of algorithms and optimiza-
tion techniques, that together, can result in order of
magnitude improvements in spatial resolution and
accuracy over standard methods. The techniques are
collectively known as Correlation Image Velocimetry
(2) To identify both the causes and consequences of the
dominant errors in practical ﬂuid mechanics applications.
The analysis is based on CIV, but the major points are
quite general, applicable to many IV methods.
(3) To outline a quantitative analysis on the feasibility of
measuring fully turbulent ﬂuid ﬂows with IV methods:
what Reynolds numbers can be reached, and at what
Analysis of errors
Experiments and simulations
A systematic approach was taken to the development of
the CIV system. Individual parameters relating to particles,
lighting, algorithms, and ﬂuid vorticity, were isolated. Exten-
sive optimization allowed their effects on the accuracy of the
technique to be determined.
Randomly distributed numerical particles with Gaussian
proﬁles were projected onto a virtual pixel array (integrating