TOWARDS A

RECONFIGURABLE
TRACKING SYSTEM
Presentation Topic for:
ICT6900: Reconfigurable Computing

Presented To: Presented By:

Dr. Md. Liakot Ali Mohammad Nazmul Haque
Associate Professor, Roll No: M10063102
IICT, BUET M. Engg. In ICT
TOWARDS A RECONFIGURABLE
TRACKING SYSTEM
 Researchers:
 Wong, S.G. Jasiunas, M. Kearney, D.
 This paper appears in:
Field Programmable Logic and Applications, 2005. In

Publication Date: 24-26 Aug. 2005

On page(s): 456- 462
ISBN: 0-7803-9362-7
INSPEC Accession Number: 8813959
Digital Object Identifier: 10.1109/FPL.2005.1515764
Current Version Published: 2005-10-10
Computer Vision
 Computer Vision is concerned with problem
of enabling a computer to see.
 computer to see = automatically (detect +
track + recognize objects) in image
sequences.
 To enable a Computer to See in image
sequences is computationally expensive.
 It must be done in real-time.
Imaging Tracking System

 Detection is a decision making process; ‘is

there an object or feature of interest at location
(i,j)?’
 Tracking is the process of associating discreet
detections over time into a track path.
 Recognition uses the results of detection and
tracking to classify the object as belonging to a
particular type.
Paper’s Contribution
 Thepaper describes methods for
implementing and approximating two
computationally demanding classes of
tracking algorithms:
A) correlation tracking and
B) probability density function (PDF) tracking,
using reconfigurable computing.
Paper’s Contribution cont..
 Both of these algorithms require use of the
related operations of correlation and
convolution.
 Both of these operations are computationally expensive.
 This paper reviews the convolution and correlation
operations and describes techniques for
implementing and approximating these operations
on a RC.
 The performance of the RC implementation is
compared to that of a serial implementation on a
microprocessor.
Correlation Tracking
 Correlation trackers work by matching a
reference image (stored in memory), with the
scene image (current sampled image produced by
the sensor).
 How Works:
 The reference image is correlated with the current frame
of the scene image to produce a correlation surface.
 The location of target object is found within scene image;
the location corresponds to the peak of the correlation
surface.
 The object is extracted from the scene image and
used to update the reference image.
Fig: A typical correlation tracker
Pros & Cons of
Correlation Tracking
 Correlation trackers can be extremely robust
at tracking objects
 Particularly useful for tracking extended
objects [8 or more pixels in area ]
 Do not require motion information to be
effective
 The correlation operation can be extremely
slow for large images.
Probability Density Function
(PDF) tracking
 PDF tracker represents the uncertainty of
the location of object in a scene as a
discrete probability density function.
 The PDF tracker is another computationally
intensive class of tracking algorithms
 The problem with tracking process is
information loss at the detection stage.
 For difficult (dim) targets this can be a problem.
PDF Tracker
 uncertainty in the position of an object to be stored
in a 2D array (an image)
 similarly velocity (and other derivatives) may be
represented by 2D arrays
 to traverse between the position and velocity
spaces, the matrix operations of correlation and
convolution may be used.
Modified alpha-beta Tracker
 Here a single target tracker, the alpha-beta tracker in PDF
form can be described by the equations:

 Where, X is position,
 V is velocity, and
 the subscripts, m, s, and p represent measured, smoothed
and predicted values.
 PDF representation of the
alpha beta tracker
 the PDF representation of the alpha beta tracker is
given by equations

 P is the positional PDF represented as an image over the space

(x,y),
 V is velocity PDF represented as an image over the space (i,j),
 ⊗ is the convolution operation,
 Ps` [n-1] is Ps[n-1] rotated by 180 degrees.
Pros Cons of PDF Tracker
 When converting a tracking filter, such as the alpha-
beta tracker, to PDF form:
 the amount of data required at each stage of the filter
increases from a single data point to a two dimensional
data-space
 to perform PDF-alpha-beta tracking requires a two-
dimensional correlation, in equation (5)
 another two-dimensional convolution in equation (6) for
every frame.
 Like the correlation tracker this is extremely
computationally expensive.
Convolution and correlation
 The convolution of two functions a and b is given by,

 If u and v represent the vector spaces (x,y) and (i,j), then a

two dimensional convolution is formed.

 For discrete images A and B this can be described

as,

 Where, A(x,y) is the input image and B(i,j) is the kernel

image with size (2m+1) by (2n+1).
Convolution and correlation
cont..
 The convolution operation is commonly used to spatially
filter images;
 performing functions such as edge detection, blurring and
sharpening.
 Convolution is related to correlation with one of the images
(typically the kernel) rotated by 180 degrees.

 where B` is B rotated by 180 degrees.

 Given that A is an image with size p by q, then to perform

convolution or correlation requires p*q*(2m+1)*(2n+1)
arithmetic operations, and a similar number of memory
accesses.
Methods for Fast Convolution
and Correlation
 Implementation Methods:
 transforming the images into other less
operationally complex domains
 mathematical and computational approximations
 parallel implementation
 combinations of these techniques
Fast Convolution (other domain)
 Transform both the image into Frequency Domain Using the
FFT (Fast Fourier Transform)
 Multiplication done in the frequency domain
 (equivalent to convolution in the spatial domain)

 Then use the inverse FFT (iFFT) to transform the result into the
spatial domain.
 this reduces the number for calculations to perform
convolution to p*q.

 There exist off-the-shelf logic cores for fast hardware

implementations of the FFT and iFFT .
 Furthermore, a 2D FFT can be constructed from 1D FFTs.
Improving
Computational Performance
 By use mathematical or computational
approximations:
 Many convolutions use symmetrical or regular shaped
masks,
 By use of an approximation of the Laplacian
mask and regular values in the mask to develop
a fast approximation to convolution
 demonstrates a 20 times speed improvement for edge
detection using a 13 by 13 pixel mask, on a
microprocessor.
Parallel Implementation on
FPGA
 Microprocessors limit the speed of convolution and
correlation
 they typically have less than four parallel hardware multipliers
available.
 A modern FPGA can have more than 150 hardware multipliers
and more than 1MB of distributed RAM
 FPGA based reconfigurable computers can drastically
improve the performance of parallel operations
 Algorithmic parallelism makes use of the nature of the
algorithm to subdivide it into separate parallel computations.
 it is possible to divide this algorithm into p*q separate parallel
operations
 Pipeline parallelism reduces the amount of slow global
memory access by exchanging them for fast local accesses,
typically using registers.
Reconfigurable
Tracking System
 The reconfigurable tracking system consists of:
 A dual Opteron workstation running at 1.7GHz
 one high speed PCI-X bus
 second independent PCI bus
 Two RC2000 PCI-X boards (ADM-XRCII) each with a 6 million gate
Xilinx Virtex-II FPGA (XCV2V6000)
 6 banks of 2 clock-cycle pipelined RAM, 2Mb per bank
 A Raytheon HS IR camera
 Bitflow RoadRunner frame-grabber PCI board

 The Opteron workstation acts as the host for the two RC2000
FPGA boards,
 acting as a benchmarking platform for comparison between
software and hardware implementations.
A simple architecture for
spatial filtering
Pipelined convolution
Pipelined convolution
Configuration for correlation
tracking
Comparative Performance of Serial
and Parallel Implementations.
Results and Conclusion
 Both correlation tracking and PDF tracking are
computationally expensive requiring multiple convolutions and
correlations every image frame.
 Serial implementations of correlation and convolution will be
too expensive for large images for the expected future.
 Reconfigurable computing offers the possibility of
implementing complex operations in parallel, with a
computational performance approaching that of custom
hardware but with the flexibility of software.
 We have demonstrated one method of rapidly performing
convolution and correlation in hardware, and used this
method to perform correlation tracking and PDF tracking.