Real-time image processing for particle tracking velocimetry
Mark Kreizer and Alex Liberzon
Turbulence Structure Laboratory, School of Mechanical Engineering, Tel Aviv University
April 6, 2009
Abstract
We present a novel high speed particle tracking velocimetry (PTV) experimental system. Its
novelty is due to the FPGA-based, real-time image processing "on camera". Instead of an image,
the camera transfers to the computer using a network card, only the relevant information of the
identi¯ed °ow tracers. Therefore, the system is ideal for the remote particle tracking systems in
research and industrial applications, while the camera can be controlled and data can be transferred
over any high-bandwidth network. We present the hardware and the open source software aspects
of the PTV experiments. The tracking results of the new experimental system has been compared
to the °ow visualization and particle image velocimetry measurements. The canonical °ow in the
central cross section of a a cubic cavity (1:1:1 aspect ratio) in our lid-driven cavity apparatus is
used for validation purposes. The downstream secondary eddy (DSE) is the sensitive portion of
this °ow and it's size was measured with increasing Reynolds number (via increasing belt velocity).
The size of DSE estimated from the °ow visualization, PIV and compressed PTV is shown to agree
within the experimental uncertainty of the methods applied.
1 Introduction
Three dimensional particle tracking velocimetry is among the state-of-the-art experimental methods
in °uid mechanics (Tropea et al., 2007; Dracos, 1996), allowing for the °uid velocity and in some
cases velocity derivatives ¯elds (LÄuthi et al., 2005; Liberzon et al., 2005) to be accurately measured
in optically transparent conditions. The main requirement of this imaging method, in which every
particle is tracked in time and space is the appropriate time resolution of the imaging device. For
very high speed °ows specially designed solutions of silicon-strip detectors (Voth et al., 1998, 2002),
or acoustic imaging devices (Mordant et al., 2004) have been developed. Most of the systems,
however, attempt to use optical imaging devices, such as CCD or CMOS digital cameras (e.g.
Dracos, 1996; Ra®el et al., 1998; Tropea et al., 2007, among others). The main limitations of the
1

optical applications for the high speed °ows lies in the high data rates that need to be streamed and
stored in real time. Therefore, the main motivation of the presented work is to develop the imaging
devices that i) easy to operate as the digital imaging device, and ii) reduces the data rates during
particle tracking velocimetry experiments. One great solution of this kind was recently proposed
by the group of Voth (see Chan et al., 2007). The authors developed a digital circuit which is
located between the camera and the frame-grabber, reducing the data rates by the factor of 100
to 1000, using threshold-based binarization algorithm. We present a di®erent solution, based on
the FPGA (¯eld-programmable gate array) on-camera-board real time image processing, that uses
Sobel ¯lter, edge detection and ¯ll-in-hole algorithm. The commercially available digital CMOS
camera (1280 £ 1024 pixels, 8 bit, 12 ¹m per pixel, Mikrotron GmbH, Germany), augmented with
the in-house software for particle tracking (http://ptvwiki.netcipia.net) was implemented and
measured the °ow in the lid-driven cavity °ow. In order to test the developed hardware/software
solution it is compared here with the °ow visualization and particle image velocimetry (PIV) results,
obtained using the same apparatus and the same experimental conditions. This is because of a
di®erent nature of the particle tracking results (sparse, scattered gridded data) versus the results
of the PIV, located at ¯xed locations. In order to quantitatively compare the experimental results,
we propose to use the size of the downstream secondary eddy (DSE), the vortical structure in the
lid-driven cavity °ow, most sensitive to the slight changes in experimental conditions. Beyond the
comparison of the experimental methods, we present an interesting phenomenon related to the size
of the DSE in the "classic" lid-driven cavity °ow, previously unattended.
A lid-driven cavity °ow is an ideal model for the °ows in industry (e.g. cutouts, slots in heat
exchanges, cavities in chemical mechanical polishing process of silicon wafers) and nature (e.g.
sediment bed impurities). It has also a signi¯cant advantage of being numerically resolved, hence
the experimental results could be compared in many aspects to the fully resolved datasets. The
most noticeable works are mentioned in the review of Shankar and Deshpande (2000), from which
we would like to emphasize the experimental works of Pan and Acrivos (1967); Kose® and Street
(1984a); Prasad and Kose® (1989), among others. Kose® and Street (1984a,b,c) investigated the
in°uences of SAR (span aspect ratio) on the size of the DSE, and partially also its variation with
the Reynolds number (ReB = UbB=º, B is the streamwise width of the cavity and Ub denotes the
bulk). Their results were based on °ow visualization photographs in a cavity of square cross section
(almost twice larger than our cavity dimensions: B = D = 150 mm).
We extend the previous studies on the size of the DSE in a square cross section (depth-wise
aspect ratio, DAR 1:1), SAR 1:1 lid-driven cavity °ow, using the quantitative experimental methods
(PIV and PTV) in addition to the qualitative °ow visualization studies. We can check if there are
di®erences between the °ow visualization and the velocity ¯eld distributions.
In agreement with experimental results of Pan and Acrivos (1967), we measure that the DSE
2