DESIGN AND DEVELOPMENT OF ENERGY
EFFICIENT COMPRESSIVE SENSING BASED
APPLICATIONS FOR WVSN
A SYNOPSIS
Submitted by
SUBBULAKSHMI T.C.
in partial fulfilment of the requirements for the degree of
DOCTOR OF PHILOSOPHY
FACULTY OF COMPUTER SCIENCE AND
ENGINEERING
ANNA UNIVERSITY
CHENNAI 600 025
OCTOBER 2018
2
1. INTRODUCTION
Due to the advancement of technology, several real-time applications
are presented to the society for improvising the quality of life. The real-time
applications are claimed to be successful only when better accuracy rates with
reasonable speed is proven. In order to improve the quality of service and the
standard of the application, several changes are incorporated to the existing
technologies.
On that front, the Wireless Sensor Networks (WSN) are tailored with
respect to the requirements of the application.
Wireless Visual Sensor
Networks (WVSN) is one among many kinds of network and it has gained
substantial research interest due to its wider applicability. Usually, the WVSN
is employed for real-time intelligent systems.
A WVSN is composed of numerous wireless nodes, which are packed
with a processor, power back-up, image sensor and transceiver [1, 2]. The
nodes of this network sense and share the data with the Base Station (BS) or
the powerful sink node, which is meant for processing the data locally.
As WVSN is based on WSN, the basic characteristics such as
distributed environment, wireless communication and energy restriction are
inherited to it. The major functionality deviation of WVSN from WSN is the
capability of environment sensing. The sensor nodes of WVSN capture the
environment in three different dimensions in the place of normal one. Hence,
WVSN is usually employed in environment monitoring and surveillance based
applications.
The sensor nodes of WVSN take the video of the specific area and
forward the captured video to the processing unit by means of relay nodes.
The underlying point here is that it is unnecessary to transmit the video to the
processing unit, unless some variation or change is involved in the video. This
idea conserves energy, which in turn helps in maximizing the lifetime of the
3
network. At this juncture, the concept of Compressive Sensing (CS) comes
into picture. The CS technique accumulates and forwards the mandatory
components (X) instead of the entire Y samples [3]. Though the complete
samples are not forwarded, the CS ensures the better data recovery. This is
achieved by computing the sparsity degree of the signal.
The sensor node of WVSN takes the video of a scene and performs
CS for computing the sampling measurements. The so computed sampling
measurements are forwarded to the destination and the video is rebuilt by the
recovery algorithm.
This research work is based on the utilization of immobile cameras,
such that the background is the same at all times. Hence, in order to conserve
energy the background part of the video can be subtracted. In other words, the
foreground of the video can easily be processed, which is the target of any
application.
Understanding the benefits of compressive sensing, this research work
intends to propose three different research solutions for WVSN that ensure
energy efficiency. The initial research phase aims to present a forepart
detection mechanism with dynamic compressive measurements. The second
research phase proposes a compressive sensing technique by considering the
texture content of the input. The final phase of this research introduces a
technique to fuse images with the help of compressive sensing.
The performances of all the proposed approaches are tested in terms
of standard performance measures and the proposed approaches yield
satisfactory results.
2. RELATED LITERATURE
This section aims to provide the related works in the existing literature
with respect to compressive sensing in WSN.
4
In [4], a parking lot occupancy detection system is proposed by
utilizing WVSN. This system analyses the video and performs compression.
This system is claimed to be flexible and can be utilized in the environment
with multiple camera nodes. A relevance based approach for multi-sink
mobility is presented for smart city application by WVSN in [5]. This work
presents an algorithm for effective localization of several mobile sinks over
the roads and streets. This approach increases the data transmission rate by
placing the mobile sink node closer to the source nodes with better sensing
relevance.
In [6], an energy efficient low bit rate image compression is presented
in wavelet domain for image sensor networks. This work proposes a new
approximation band transform algorithm and it works by extracting and
encoding the image approximations by means of fixed-point arithmetic. This
idea conserves the energy of resource constrained sensor networks. The
sensing, coding and transmission functionalities of visual sensor networks for
smart city applications are modelled by means of fuzzy based approach in [7].
This work configures the activities of the sensors such as sensing, coding and
transmission in a dynamic way by utilizing different parameters.
An energy efficient image compressive transmission scheme is
proposed for wireless camera networks in [8]. This work states that this work
can extract the regions of interest, controls the quality of image and carries out
better image communication. The performance of this approach is tested in
terms of image quality, execution time and energy consumption. In [9], an
energy efficient image transmission scheme for wireless multimedia sensor
networks is presented on the basis of block based compressive sensing. The
encoding algorithm of CS measurements is proposed by considering Bernoulli
measurement matrix. The performance of the proposed approach is tested in
terms of energy consumption and image quality.
5
In [10], an image coding scheme based on compressive sampling is
presented for visual communication. This work encodes the image by
performing polyphase down-sampling. Additionally, the local random
convolutional kernel is utilized before the process of down-sampling. These
measurements are utilized for preserving the features with greater frequency
and to remove redundancy.
An efficient image coding and transmission system based on scrambled
block compressive sensing is presented in [11]. This work employs scrambled
block compressive sampling for performing image measurement by means of
a sensing operator. This is followed by the execution of progressive nonuniform quantization and the progressive non-local low-rank reconstruction is
utilized at the decoder side.
In [12], a Discrete Cosine Transform (DCT) based multi-focus image
fusion scheme is presented for VSN. The fusion system mainly focuses on the
Alternating Current (AC) coefficients computed in the DCT domain. This
method is claimed to prove better image quality and minimal energy
consumption.
A clustering based data compression scheme based on k-means
algorithm is proposed for wireless imaging sensor networks in [13]. This work
compresses the images by means of k-means algorithm by considering the
colour of image pixels and the image is compressed. A survey on sensor
coverage, visual data acquisition, processing and transmission is presented in
[14].
An adaptive compressed sensing rate assigning algorithm based on
standard deviation of the image blocks for wireless image sensor network is
presented in [15]. In this work, all the image blocks are fixed with a static
sampling rate. Additionally, an adaptive sampling rate is passed into all the
blocks, such that greater sampling rates are allotted to blocks that are minimal
6
compressible. Finally, the fixed and the adaptive measurements are considered
together to form the final measurements.
In [16], an adaptive compressive sensing for tracking the targets in
WVSN based surveillance is presented. The adaptive compressive sensing can
achieve greater compression rates with respect to the sparsity nature of
different datasets. In addition to this, it is unnecessary to compress the whole
image; instead the relative blocks that contain the target can alone be
compressed. The image compression algorithms for wireless multimedia
sensor networks are reviewed in [17].
In [18], a video compressed sensing framework is presented for
wireless multimedia sensor networks by utilizing a combination of several
matrices. This work utilizes the combination of Discrete Wavelet Transform
(DWT) and Discrete Cosine Transform (DCT) for performing compression.
This work is proven to be better than Gaussian matrix.
3. RESEARCH MOTIVATION
Due to the advent of modern automation technology, the WSN
flourishes with several interesting branches. WVSN is one of the important
branches, which attempts to capture videos from different places and it plays
an important role in the area of security. As the WVSN deals with real-time
videos, the energy of the sensors must be utilized wisely.
Considering this fact, this research work pays utmost attention to bring
in energy efficiency by incorporating the concept of compressive sensing in all
the three research phases. Each phase is unique and the overall theme of this
work is to ensure energy efficiency. The energy efficiency of the sensors
results in improvised lifespan of the network.
7
4. RESEARCH GOALS
The central goal of this research is to attain energy efficiency in
WVSN, such that the lifespan of the network is reasonable and purposeful. As
WVSN deals with multimedia data, the energy of the sensors drains faster.
The energy consumption of the sensors is balanced and controlled with the
help of compressive sensing.
The aim of compressive sensing is to control the amount of data
transfer, which contributes a lot in conserving energy efficiency. All the three
phases of the research work employs compressive sensing to manage the
energy consumption and proves better energy conservation.
5. ALGORITHM AND METHODOLOGY
The initial research phase presents an energy conserving scheme that
extracts the forepart from the background scene. The compressive
measurements are dynamically computed by this work, which reduces the
overhead and energy consumption. This work selects minimal yet, optimal
compressive measurements and forwards it to the destination side. The
destination side rebuilds it with the help of Compressive Sampling Matching
Pursuit (CoSaMP) algorithm.
The second phase of this research intends to present an adaptable
compressive sensing scheme by considering the textural pattern of the input
frame. This idea overthrows the concept of standardized compressive sensing,
which treats all the parts of the frame similarly. However, the adaptable
compression sensing scheme considers the texture pattern of the frame and
adjusts itself. This idea guarantees better image quality after the process of
rebuilding. The overall algorithm of this research work is summarized and
presented as follows.
8
Overall Algorithm
//Research Phase I
Input : Videos
Output : Forepart detection
Begin
𝑓𝑔
Assess 𝑘𝑡𝑝 from the compressive measurements;
Assess cross validation measurements;
Detect forepart;
If (𝑝𝑡 ≥ 𝑠𝑡𝑡𝑝 )
Transmit to the destination;
Else
Discard the measurements;
End if;
// Destination
Rebuild by CoSaMP;
End;
//Research Phase II
Input :Videos
Output : Adaptable compressive sensing
Begin
Sample the input video frame;
Calculate the texture by GLVP;
Fix the sampling rate;
Construct the transformation matrix mat;
Process mat and rebuild the image by CoSaMP;
End;
//Research Phase III
Input : Videos from different sensors
Output : Detailed images
Begin
Extract video frames;
9
For all video frames
do
Apply Shearlet and Curvelet over the image frames;
Compute local spatial frequency row-wise and column-wise;
Fuse the video frames;
End do;
End;
The third phase of the research attempts to fuse the images being
captured by different sensors based on compressive sensing, so as to gain
more information about the image. This concept of image fusion results in
better visibility of the image by providing intricate details. Hence, all the three
phases employ compressive sensing for achieving better energy efficiency,
while preserving the quality of the frame.
The performance of the proposed approaches is validated by means of
standard performance measures such as energy efficiency and quality based
measures such as Peak-Signal-to-Noise-Ratio (PSNR) and Mean Square Error
(MSE).
6. RESULTS
The proposed approach is simulated in Matlab environment on a stand
alone computer with 8 GB RAM. The proposed approach is analysed by
utilizing different video streams downloaded randomly. The size of the video
frame is 288 × 352, the sparsity and the compressive measurements are
dynamic. The sample results of the proposed approaches are tabulated as
follows.
10
Table 1: Experimental Results of Proposed Approaches
Research Phase 1 – Energy Conserving Forepart Detection Scheme with Dynamic
Compressive Measurements based on Compressive Sensing for WVSN
Comparative Techniques
Performance measures
Precision
Recall (%)
F-Measure
Average Energy
(%)
(%)
Consumption
Rate (mJ)
CS-BS [19]
37.8
38.2
37.99
68.56
CS-BS [20]
77.2
75.8
76.49
65.26
Hybrid Matrix Based [21]
87.9
84.4
86.11
8.3
Proposed
89.6
86.7
88.12
7.33
Research Phase 2 – Texture based Adaptable Compressive Sensing Scheme for WVSN
Comparative Techniques
PSNR
Energy
Execution Time (ms)
Consumption Rate
(mJ)
Hybrid matrix based [21]
30.83
11.8
1943
Phase I
32.98
9.9
1876
Proposed
37.8
9.2
1765
Research Phase 3 – Compressive sensing based Image Fusion for WVSN
Comparative Techniques
PSNR
Energy
Execution Time (ms)
Consumption Rate
(mJ)
Shearlet only
23.83
1897
8.9
Curvelet only
32.98
9.2
1869
Shearlet + Curvelet
10.1
1949
38.4
The above presented table presents the sample results of the proposed
approaches. From the experimental results, it is evident that the proposed
approaches perform well with better quality, while consuming reasonable
energy.
7. ORGANIZATION OF THE THESIS
The whole thesis is categorized into six chapters and the outline of each
and every chapter is presented below.
Chapter 1 presents the basic idea of WSN that includes the architecture
and branches of WSN. The fundamental concept of WVSN and the challenges
associated with it are explained. In addition to this, the motivation and
objectives of the thesis are explored in this chapter.
11
Chapter 2 intends to study the related literature with respect to the
energy efficiency of WVSN. Additionally, the state-of-the-art approaches are
studied in this chapter.
Chapter 3 presents an energy conserving forepart detection scheme with
dynamic compressive measurements by employing compressive sensing for
WVSN. The proposed approach is presented in detail and the performance of
the work is analysed and compared with the existing approaches.
Chapter 4 proposes a texture based Adaptable Compressive Sensing
Scheme for WVSN. This work considers the texture of the video frames into
account and the sampling rate is altered accordingly. This idea enhances the
quality and is proven by the PSNR value.
Chapter 5 introduces a compressive sensing based Image Fusion for
WVSN, which attempts to fuse video frames captured by infrared and visible
sensors. This work ensures better quality and preserves the energy as well.
Chapter 6 presents the concluding remarks, which includes both
positive and negative sides of the research. Additionally, the possible future
research directions are also presented.
References
[1] Tavli, B., Bicakci, K., Zilan, R. and Barcelo-Ordinas, J.M., 2012. A survey
of visual sensor network platforms. Multimedia Tools and Applications, 60(3),
pp.689-726.
[2] Soro, S. and Heinzelman, W., 2009. A survey of visual sensor networks.
Advances in multimedia, 2009.
[3] Donoho DL. Compressed sensing. IEEE Trans Inform Theory
2006;52:1289–306.
12
[4] Baroffio L, Bondi L, Cesana M, Redondi AE, Tagliasacchi M. A visual
sensor network for parking lot occupancy detection in smart cities. InInternet
of Things (WF-IoT), 2015 IEEE 2nd World Forum on 2015 Dec 14 (pp. 745750). IEEE.
[5] Peixoto JP, Costa DG. Wireless visual sensor networks for smart city
applications: A relevance-based approach for multiple sinks mobility. Future
Generation Computer Systems. 2017 Nov 1;76:51-62.
[6] Phamila AV, Amutha R. Energy-efficient low bit rate image compression
in wavelet domain for wireless image sensor networks. Electronics Letters.
2015 May 14;51(11):824-6.
[7] Costa DG, Collotta M, Pau G, Duran-Faundez C. A fuzzy-based approach
for sensing, coding and transmission configuration of visual sensors in smart
city applications. Sensors. 2017 Jan 5;17(1):93.
[8] Wang Y, Wang D, Zhang X, Chen J, Li Y. Energy-efficient image
compressive transmission for wireless camera networks. IEEE Sensors J..
2016 Mar;16(10):3875-86.
[9] Hemalatha R, Radha S, Sudharsan S. Energy-efficient image transmission
in wireless multimedia sensor networks using block-based compressive
sensing. Computers & Electrical Engineering. 2015 May 1;44:67-79.
[10] Liu X, Zhai D, Zhou J, Zhang X, Zhao D, Gao W. Compressive
sampling-based image coding for resource-deficient visual communication.
IEEE Transactions on Image Processing. 2016 Jun;25(6):2844-55.
[11] Chen Z, Hou X, Qian X, Gong C. Efficient and robust image coding and
transmission based on scrambled block compressive sensing. IEEE
Transactions on Multimedia. 2018 Jul;20(7).
[12] Phamila YA, Amutha R. Discrete Cosine Transform based fusion of
multi-focus images for visual sensor networks. Signal Processing. 2014 Feb
1;95:161-70.
13
[13] Paek J, Ko J. K-Means clustering-based data compression scheme for
wireless imaging sensor networks. IEEE Systems Journal. 2015 Oct 30.
[14] Yap FG, Yen HH. A survey on sensor coverage and visual data
capturing/processing/transmission in wireless visual sensor networks. Sensors.
2014 Feb 20;14(2):3506-27.
[15] Zhang J, Xiang Q, Yin Y, Chen C, Luo X. Adaptive compressed sensing
for wireless image sensor networks. Multimedia Tools and Applications. 2017
Feb 1;76(3):4227-42.
[16] Fayed S, Youssef SM, El-Helw A, Patwary M, Moniri M. Adaptive
compressive sensing for target tracking within wireless visual sensor
networks-based surveillance applications. Multimedia Tools and Applications.
2016 Jun 1;75(11):6347-71.
[17] ZainEldin H, Elhosseini MA, Ali HA. Image compression algorithms in
wireless multimedia sensor networks: A survey. Ain Shams Engineering
Journal. 2015 Jun 1;6(2):481-90.
[18] Nandhini SA, Sankararajan R, Rajendiran K. Video compressed sensing
framework for wireless multimedia sensor networks using a combination of
multiple matrices. Computers & Electrical Engineering. 2015 May 1;44:51-66.
[19] Cevher Volkan, Sankaranarayanan Aswin, Duarte Marco F, Reddy
Dikpal, Baraniuk Richard G, Chellappa Rama. Compressive sensing for
background subtraction. In: Computer vision-ECCV 2008. Berlin Heidelberg:
Springer; 2008. p. 155–68.
[20] Manimozhi S, Aasha Nandhini S, Radha S. Compressed sensing based
background subtraction for object detection in WSN. In: IEEE international
conference on communication and signal processing (ICCSP).
[21] Sukumaran AN, Sankararajan R, Swaminathan M. Compressed sensing
based foreground detection vector for object detection in wireless visual
sensor
networks.
AEU-International
Communications. 2017 Feb 1;72:216-24.
Journal
of
Electronics
and
14
Research Publications
1. Subbulakshmi T.C., Dr. Gnanadurai, D., Dr. Muthulakshmi I., 2018,
“Energy Conserving Forepart Detection Scheme with Dynamic Compressive
Measurements based on Compressive Sensing for WVSN”, Journal of Internet
Technology, Accepted.
2. Subbulakshmi T.C., Dr. Gnanadurai, D., Dr. Muthulakshmi I., 2018,
“Texture based Adaptable Compressive Sensing Scheme for WVSN”,
International Journal Intelligent Engineering and Systems, Article in Review.
3. Subbulakshmi T.C., Dr. Gnanadurai, D., Dr. Muthulakshmi I., 2018,
“Compressive sensing based Image Fusion for WVSN”, International Journal
Engineering and Technology (UAE), Article in Review.