Initiation Into Research Fondecyt Project: Optimal
packetization of still images in wireless vision sensor networks
Abstract
Wireless visual sensor networks (WVSNs) represent a technological
challenge that multiplies the difficulties of the well known classical
wireless sensor networks (WSNs) applications. Considering the
resource-constrained nature of current sensor nodes, the large amount
of data necessary to code a single image increases the necessity of
highly efficient processing and transmission algorithms. In this
matter, significant progress has been made in the design of efficient
algorithms for image or video compression, e.g., by reducing the
computational complexity required to calculate a classic DCT . Other
works point to the improvement of communication protocols.
Nevertheless, many works put aside one of the most inherent issues in
WSNs: packet loss.
The problem of packet losses can be addressed whether through the use
of ARQ or FEC-based protocols, but at the expense of additional energy
consumptions and latency. Luckily, in the case of still images data
have a high loss-tolerance due to the correlation between neighboring
pixels. This feature is widely exploited by lossy compression
algorithms, but it can also be exploited to reinforce a transmission
scheme. One of the simplest techniques for image robust transmission is
block interleaving.
The effectiveness of the application of interleaving methods is
assumed. However, methods are generic and the choice of the correct
parameters is arbitrary. Also, there is no basis for comparison, not
knowing a standard optimum to reach when designing an interleaving
method. Characteristics of the channel, batteries level and images
(among other factors) are also not considered.
In this project, we address the issue of block interleaving techniques
studying the possibility of finding optimal interleaving configurations
and to numerically evaluate existing and future proposals. We will
focus on five problems. First, we will study error concealment and
image quality metrics the more used in current image processing.
Second, we will rigorous study block interleaving methods, analyzing
blocks' paths and parameters impact through extensive simulations. MPI
and CUDA based programming will be applied. Simulations must consider
multiple image sets and loss patterns. Comparisons, including image
quality of reconstructed images and computing cost, will be provided.
Third, we will propose one (or various) objective functions, being a
representative reference for evaluating the performances of an
interleaving method. Thus, researchers will be able to estimate and
compare the efficiency of interleaving methods without the necessity of
making extensive simulations (what is mandatory for evaluations, until
now). The idea is to count on a statistically-representative objective
function to maximize (or minimize), whose validity will be supported
with comparisons with our extensive simulations.
Fourth, by using the proposed objective function, we will find optimal
parameters for existing interleaving methods, to finally (Fifth) work
on an optimal reference interleaving configuration. Considering the
proposed function, to find optimal (or pseudo-optimal) configurations
which can serve as a comparison reference for more traditional
interleaving methods. In this way, authors working on interleaving
methods will know if their proposals are near a reference optimal.
Goals
The general goals of this research project are to propose a correct
optimization function for the evaluation of block interleaving
techniques and to find optimal configurations of coded/non-coded
bitmaps, using the optimization function, to serve as a basis of
reference for further proposals on the matter.
Specific goals of this project are:
- To analyze interleaving methods by performing extensive
simulations using parallel programming with MPI and CUDA. Simulations
must be performed over big sets of images and loss patterns.
- To establish an objective function allowing the evaluation of
interleaving methods without the necessity of perform extensive
simulations and to prove that this objective function is statistically
representative.
- To study, with the proposed objective function, interleaving
methods in the bibliography, finding optimal parameters to each of them.
- To find optimal bitmap configurations, in order to count on a comparison reference on the design of interleaving methods.
Methodology
The adopted methodology will consider the following general steps:
- Extensive analysis of the literature, defining the
state-of-the-art in the matter of interleaving methods for robust image
communication, with special attention to works applied to wireless
vision sensor networks.
- Preparation of a simulation testbed considering large sets of
different images and loss patterns. Open-source application Sim-LIT
will be used as basis for software application. The simulations of this
project should be executed in a high-performance computer using
parallel-programming with MPI libraries CUDA.
- Formal modeling of possible objective functions. Objective
functions will be used with studied interleaving methods with multiple
parameters. Resulting evaluation metrics will be stored and then
compared with results of simulation for same interleaving methods with
same parameters. Statistical metrics will determine if results of the
proposed objective functions and simulations are correlated and if they
can be used to replace extensive simulations for evaluation and compare.
- To use selected objective functions to evaluate literature's
interleaving methods and to find optimal parameters for various image
resolutions and/or characteristics.
- To find optimal interleaved bitmaps, using the selected objective
functions, using genetic algorithms, simulated-annealing, and other
search algorithms. Then, to compare optimal values obtained from
previous applications and to compare with optimal bitmaps in order to
find out if optimal bitmaps are better or (at least) equal to them.
Results of followed steps will be constantly documented. It is expected
that this documentation becomes basis for publications in research
journals and national/international conferences.
During the development of the present project, it is expected to share
results and keep research discussions with Dr. Vincent Lecuire, from
the Research Centre for Automatic Control of Nancy (CRAN), France.
In terms of the use of computing resources, all of the development and
documentation will be carried out with open software, mainly based on
distributions of the operating system GNU Linux and C++-based
programming. All of the developed software will be correctly documented
and codes published for open access to research community.
Project documentation will be written in LaTeX and there will be a
database with documentation and reference will be registered in a .bib
file. All of the results will be always stored in separated files with
resumed explanations and reconstructed images correctly labeled with
significant names.
Poster:
- Project poster. Presented in III UBB Research Meeting (III
Encuentro de Investigación UBB). Universidad del Bío-Bío, Chillán,
Chile. Jan / 10-11 / 2013. [poster (pdf | en)][poster (jpg | en)]
Publications
- Katherine Zapata-Quiñones, Cristian Duran-Faundez, and Gilberto Gutiérrez. "Algoritmo Genético para la Generación de Configuraciones Óptimas de Paquetización de Imágenes". In: XXVII Encuentro Chileno de Computación (ECC 2015), Santiago, Chile, November, 2015. [slides (pdf | sp)]
- Joaquin Alder Silva-Faundez, Cristian Duran-Faundez, Pedro Melin, and Cristhian Aguilera. "An anticipatory control for a flexible manufacturing system based
on the perception of mobile units using wireless sensor networks". In: International Journal of Computer, Communications & Control, 2015, 10(5), 702-717. DOI: http://dx.doi.org/10.15837/ijccc.2015.5.634.
- Cristian Duran-Faundez, Jonathan Matías Palma Olate, Eric Orellana-Romero, and Pedro E. Melin. Effect of packet sampling time on a colony of mobile routing robots for communication link maintenance using IEEE 802.15.4 devices. In: 10th International Conference on MOdeling, Optimization and SIMlation (MOSIM 2014),
Nancy, France, November 2014.
- Eric Orellana-Romero, Javier SanMartin-Hernandez, Cristian Duran-Faundez, Vincent Lecuire, and
Katherine Zapata-Quinones. Evaluation of Block Interleaving Techniques for Robust Image Communication in Wireless Camera
Sensor Networks. In: 2014 IEEE Conference on Wireless Sensors (ICWiSe 2014),
Kuala Lumpur, Malaysia, October 2014.
People:
- Cristian Duran-Faundez (DIEE - UBB, Chile)
- Vincent Lecuire (CRAN - U. Lorraine, France)
- Eric Orellana-Romero (MCC - UBB, Chile)
- Katherine Zapata (MCC - UBB, Chile)
Financial support
[Home]
Last
modified: Mar-28-2016