RESEARCH ARTICLE
Three-dimensional particle tracking method using FPGA-based
real-time image processing and four-view image splitter
Mark Kreizer • Alex Liberzon
Received: 31 March 2010 / Revised: 4 August 2010 / Accepted: 13 August 2010 / Published online: 26 August 2010
 Springer-Verlag 2010
Abstract We present a cost-effective solution of the
three-dimensional particle tracking velocimetry (3D-PTV)
system based on the real-time image processing method
(Kreizer et al. Exp Fluids 48:105–110, 2010) and a four-
view image splitter. The image processing algorithm, based
on the intensity threshold and intensity gradients estimated
using the fixed-size Sobel kernel, is implemented on the
field-programmable gate array integrated into the camera
electronics. It enables extracting positions of tracked
objects, such as tracers or large particles, in real time. The
second major component of this system is a four-view
split-screen device that provides four views of the obser-
vation volume from different angles. An open-source ray-
tracing software package allows for a customized optical
setup for the given experimental settings of working dis-
tances and camera parameters. The specific design enables
tracking in larger observation volumes when compared to
the designs published up to date. The present cost-effective
solution is complemented with open-source particle track-
ing software that receives raw data acquired by the real-
time image processing system and returns trajectories of
the identified particles. The combination of these compo-
nents simplifies the 3D-PTV technique by reducing the size
and increasing recording speed and storage capabilities.
The system is capable to track a multitude of particles at
high speed and stream the data over the computer network.
The system can provide a solution for the remotely con-
trolled tracking experiments, such as in microgravity,
underwater or in applications with harsh experimental
conditions.
1 Introduction
Lagrangian description of turbulent flows attracts increas-
ing attention due to its simple conceptual basis and direct
relation to the studies of transport, mixing and dispersion
(Toschi and Bodenschatz 2009). Optical tracking of small
particles, bubbles or droplets, is one of the major tools used
to obtain Lagrangian trajectories for turbulence research
Dracos (1996); Lu¨thi et al. (2005). With continuously
increasing recording speed of imaging devices, the limi-
tations have shifted to streaming and storage of the high-
speed video data. High-speed digital cameras can record
video at few gigabytes per second, but cannot transfer more
than hundreds of megabytes per second to the storage
device. A possible solution is a real-time compression, e.g.
Chan et al. (2007) that reduces the data rate by a factor of
100–1000. For the tracking of sparse small objects, we
proposed to decrease the data rate differently, implement-
ing real-time image processing algorithm on the field-
programmable gate array (FPGA) (Kreizer et al. 2010).
The system can detect and localize a thousand particles in
real time. The main advantage of the FPGA integrated in
the camera electronics, rather than on an external device, is
the strongly reduced data transfer rate. This enables oper-
ation of the remotely controlled camera using the existing
network protocols and devices and significantly improves
the mobility of particle tracking systems. In the present
study, we extend the system for the three-dimensional
particle tracking at high speed, instrumenting a real-time
M. Kreizer  A. Liberzon (&)
Turbulence Structure Laboratory,
School of Mechanical Engineering,
Tel Aviv University, Ramat Aviv 69978, Israel
e-mail: alexlib@eng.tau.ac.il
M. Kreizer  A. Liberzon
International Collaboration for Turbulence Research (ICTR),
Goettingen, Germany
123
Exp Fluids (2011) 50:613–620
DOI 10.1007/s00348-010-0964-3

image processing camera tracking with a four-view split-
screen optical device. Similar optical designs have been
used in a single camera stereoscopic particle image ve-
locimetry (PIV) (e.g. Arroyo and Greated 1991; e.g. Bardet
et al. 2010), and particle tracking experiments (e.g. Tsorng
et al. 2006; Hoyer et al. 2005; Guala et al. 2008) used a
multi-mirror optical setup to obtain three-dimensional
particle tracking. The present optical setup has slightly
different design, aimed at tracking of particles in larger
observation volumes, when compared to the splitter design
of Hoyer et al. (2005). The presented remotely controlled
single-camera 3D particle tracking system can be useful for
the large-scale tracking experiments (Lobutova et al.
2009), microgravity experiments (Dupont et al. 1999), or
underwater applications (Nimmo Smith 2008). The system
enables high-speed image recording rates and does not
require synchronization of multiple cameras. However, due
to a smaller number of pixels per view, one has to com-
promise the lower particle seeding density, when compared
to the multi-camera systems.
The paper is organized as follows. Sect. 2 describes in
detail the optical design and the three-dimensional particle
tracking experiment setup. We report on the quality tests
and preliminary results in Sect. 3 and summarize it in
Sect. 4.
2 Experimental method
2.1 Experimental setup
We use a lid-driven cavity flow facility, described in detail
in Kreizer et al. (2010). The flow facility and the experi-
mental and optical setups are shown schematically in
Fig. 1. The flow in the cavity (1) is driven by a belt (2) that
slides above the glass tank. The speed of the belt is con-
trolled by the computer-managed AC motor. We use a
high-speed digital CMOS camera of 1280 9 1024 pixels
@ 500 frames-per-second with a pixel size of 12 lm
(MC1324, Mikrotron GmbH, Germany), with an integrated
FPGA. The camera can transfer raw images or processed
data through an Ethernet cable to a host PC, using GigE
protocol. The processed data contains the positions of the
center-of-mass and intensity of the particles in image
coordinates, obtained from the x and y pixel coordinates of
the left/right/top/bottom edges, the sum of the intensity
values in horizontal and vertical direction and total inten-
sity of the original particle. Details of the image processing
algorithm were presented in Kreizer et al. (2010), along
with the two-dimensional tracking and PIV results in the
lid-driven cavity flow.
The main purpose of the new development is to develop
a 3D tracking system that will complement the existing
high-speed 3D-PTV system for the two-phase flow exper-
iment similar to Guala et al. (2008). In the two-phase flow
experiment, the existing multi-camera system could track
densely seeded flow tracers, whereas the present system
will track the large solid particles. Therefore, as a test case,
we present particle tracking of relatively large, neutrally
buoyant polystyrene particles [dp&700 lm, qp = 1.03
g cm-3, St* O(0.1)] . This experiment is better suitable
for the single-camera system which, due to the reduced
number of pixels per view, requires lower volume fraction
of the seeding material.
2.2 Optical setup
Our focus is on the extension that enables the three-
dimensional Lagrangian particle tracking. We designed a
four-view split-screen optical device (denoted as (3) in
Fig. 1a and shown in orthogonal views in Fig. 1b), here-
inafter called ‘‘image splitter’’, and an array of back-side
mirrors such that the camera records four views of the
cavity, as shown in the top view of Fig. 1b.
(b)(a) Front view
Top view
Fig. 1 a Schematic view of the
lid-driven flow facility and the
experimental setup. (1) cavity,
(2) driving belt system, (3) four-
view image splitter , (4) back-
side mirrors, (5) the camera.
b Orthogonal views of the four-
view image splitter. Shaded
region denote the useful
imaging area
614 Exp Fluids (2011) 50:613–620
123