Parallel Image Processing On A Massively Parallel Pyramid Architecture For Robotic Vision Systems
Loading...
Date
1994-11
Authors
M Rabti, Al-Asaad
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Robots will not be able to do useful work until they can wander within and interact
with dynamic environments. To do that they must be able to recognize and perceive realworld
objects as they move about and change in real-time. F.,o r real-time vision, objects
must be recognized within milliseconds. Real-time image processing and analysis is
computationally intensive, and it requires the use of massively parallel machines.
Conventional parallel machines consist of an array of identical processors organized in
either single instruction multiple data (SIMD) or multiple instruction multiple data
(MIMD) configurations. Machines of this type generally only operate effectively on parts
of the image analysis problem. SIMD operates on low-level processing and MIMD
operates on high-level processing. This thesis considers the implementation and the
development of real-time parallel image processing algorithms for use in robotic- vision
systems on a massively parallel pyramid machine, called SPHINX. This machine
consists of both SIMD and MIMD parts in a multiple-SIMD organization which can
operate effectively at all levels of the image analysis problem. It can provide fast local and
global properties compu41tion of an image. The complexity of an operation is reduced by
O(210gn), where (n x n) is the input image size. A successful general-purpose simulator
based on the prototype SPHINX pyramid machine has been upgraded and realized, using
*Lisp data parallel language. Thus, the functionality of the parallel algorithms developed
for "real-time" execution is tested. A number of parallel common vision algorithms has
been implemented and developed on the pyramid machine. These algorithms address the
stages of enhancement, edge detection, segmentation, and detection of straight lines and
circles. Their performance is analyzed using the characteristics of the actual pyramid
machine. They demonstrate the power of the multi-SIMD architecture for performing
image analysis in real-time. The Hough transform is a well-known medium-level image
recognition technique, for the detection of straight lines and circles. Its adoption has been
slow due to its computational and storage complexity. Various parallel algorithms based
on its technique are developed and implemented on the pyramid machine, to obtain higher
speed. Since the conventional Hou~h transform technique requires" floating point
arithmetic which is time consuming, a new efficient scheme for straight lines detection is
designed and implemented to reduce the computation time. It u.,s es only integer arithmetic.
Results on the analytic and empirical performance of these algorithms, and a comparison
of their performances and computation times are presented. The final result of this work
is a user friendly, modular parallel digital image processing package, named parallel
ALIRA. It combines "the SPHINX pyramid machine simulator, the high performance
parallel algorithms, and the graphical user interface. It is potentially capable of
performing in real-time many tasks relevant to image analysis.
Description
Keywords
Parallel Pyramid Architecture , For Robotic Vision Systems