Enhancing the lifespan of visual sensor networks using a

Enhancing the Lifespan of Visual Sensor Networks using a Predetermined Node Deployment Strategy Subir Halder and Amrita Ghosal Department of CSE, Dr. B. C. Roy Engineering College, Durgapur, India [email protected], [email protected]

Abstract— Visual sensor networks are widely used for surveillance, environmental monitoring and tracking. However, due to the funneling effect, visual sensor nodes (SNs) close to the sink tend to deplete their energy sooner than SNs farther away from the sink resulting in energy hole problem. This energy hole problem significantly increases the unbalanced energy usage among SNs and leads to premature decrease in network lifetime. Initially, we have analyzed network lifetime and found number of relay nodes and their location has significant influence on limiting the energy hole problem and enhancement of network lifetime. Based on the principle of energy balancing, derived from this analysis, we have developed a heterogeneous SNs deployment strategy leading to an enhancement of network lifetime. Exhaustive simulation is performed to substantiate our claim of energy balancing and subsequent enhancement of network lifetime. Finally, the scheme is compared with one existing deployment scheme. The results confirm our scheme’s supremacy over the existing scheme in terms of performance metrics such as energy balancing and network lifetime. Keywords—Node deployment; Energy hole problem; Energy balance; Network lifetime; Visual sensor network.

I.

INTRODUCTION

The relentless pace of technological augmentation in image sensors led to the emergence of a new ubiquitous paradigm— visual sensor network (VSN) [1]. VSNs consist of tiny SNs called camera nodes, which integrate the image sensor, embedded processor and wireless transceiver [1]. Images are captured and extracted by the camera nodes. The embedded processor performs further image processing operations such as image compression or even image characterization to make decisions based on the images. This processed data is then finally sent to the appropriate destination in the network, which forward the images to the sink or base station. Since VSNs are battery operated and require a lot of energy to capture and process image compared to wireless sensor network (WSN), as a result, the VSN as a whole must minimize the energy usage in order to enable untethered and unattended operation for an extended period of time. Therefore, a critical consideration in designing such VSNs is conserving energy to maximize the post-deployment network lifetime. The rate of energy depletion in the network primarily depends on the deployment nature of the SNs which further mainly depends on the application environment [2]. As the SNs are expensive and their operations are significantly affected by their positions so, in VSNs location-wise predetermined deployment is generally more preferable compared to random deployment.

Most of the prior research on VSN, have primarily focused on energy conservation issue for a given specific scenario. One important way of energy conservation by limiting energy hole is through balanced energy consumption. Unbalanced energy consumption in any part of VSN may stop functioning of that part leading to a phenomenon known as energy hole problem [3]. The energy hole generally occurs near to the sink due to the funneling effect i.e., the nodes closer to the sink have to forward more packets than the nodes at the periphery of the network. The energy hole problem has worse impact on a VSN than a conventional WSN. While energy consumption is dominated only by data transmission and reception in a WSN, VSN consumes extra power on image sensing, processing and storage operations. Because of such additional energy consumption, the energy hole problem inherent in WSNs has a drastic impact on the network lifetime and leads to premature decrease in network lifetime of VSN. To circumvent this, care should be taken that SNs are deployed in such a manner that energy dissipation of all SNs takes place uniformly ensuring load balancing throughout the network. Since SNs are expensive, deploying more redundant SNs closer to the sink only for energy backup does not seem to be a cost efficient option. Many works have addressed a number of issues in VSNs. However, there are still many challenges in the area of energy management [4]. In one such work, Soro and Heinzelman [5] presented two camera selection methods to avoid redundant data and conserve SNs energy. The first method is based on minimum angle between the user’s desired view direction and the camera’s direction. The second method is based on cost metrics that measure the camera’s importance to preservation of full coverage of the 3D monitored space. Yildiz et al. [6] have addressed the problem of SNs deployment in heterogeneous VSNs to provide video panorama with a trade-off between cost and resolution. Li et al. [7] have proposed a two-tier heterogeneous WSN deployment strategy to prolong network lifetime of a VSN while minimizing cost and limiting the effects of energy hole problem. Hooshmand et al. [8] have proposed SNs selection algorithm to maintain complete visual coverage of the designated aerial territory for the longest period of time. In most of the existing VSN energy management works, the proposed deployment strategies have guaranteed the increase of network lifetime. However, most of these deployment strategies do not belong to the category of location-wise predetermined deployment strategy. Here, by location-wise predetermined strategy we mean not only the number of SNs to be deployed in a network area is

determined a priori, but the precise locations are also predetermined. Most of the existing works are silent about the exact locations of placing the SNs, which is an important criterion for applications like smart surveillance of public or remote areas, smart homes for babies etc. Keeping this in mind, we analyzed the method of controlling network lifetime while balancing energy consumption considering heterogeneous SNs deployed in terms of performing tasks (sensing and relaying). Based on the analysis, energy balancing principle is derived which in turn is used for deciding the number of heterogeneous SNs to be deployed along with their precise deployment locations. We propose a heterogeneous SNs deployment scheme that achieves energy balancing to a greater extent. Performance of the scheme is evaluated through quantitative analysis. In quantitative analysis both ideal and realistic scenarios are provided for showing the impact of routing and medium access control (MAC) protocols on the performance of the strategy. The rest of the paper is organized as follows. The network architecture considered for the present work is presented in section II. Analysis on network lifetime is done in section III. A heterogeneous SNs deployment scheme exploiting the derived principle of energy balancing is proposed in section IV. In section V, the performance of the scheme is evaluated by providing simulation results. Finally, the paper is concluded with some mention about the future scope of the work in section VI. II.

SYSTEM MODEL

In this section, we describe the network architecture along with the necessary assumptions. This section also presents communication and sensing models, energy model and the concept of network lifetime considered for this work.

L4

L3 C13 L2

C33

C32 C 22

C12

L1 C11

rc

C 24

C14

C34

C 44

C34

C53 B1

C12

C74

C64

C54 C36

rs

C37

C84 C83

b Strategic location

C82

r

C18

a Sink

Fig. 1. Regular hexagonal cell based network architecture.

A. Network Model Similar to the models in [3], we assume that all SNs are distributed in a two dimensional plane covered by a set of regular hexagonal cells (RHCs) as shown in Fig. 1. The network coverage area (a×b) is divided into N number of layers and each layer is divided into M number of RHCs with radius r. We identify a layer as Li where i=1,2,…,N. A layer is called either odd or even based on the value of i. For example, L2 , L4 ," are the even layers whereas L1 , L3 ," are the odd layers (refer to Fig. 1). Further, the sink is assumed to be located at one corner of the network area and responsible for collecting data from the nodes. A RHC in the network is identified by Cij (where i=1,2,…,N and for each i, j=1,2,…,M) which indicates the jth cell in layer-i. For

example, the RHC C24 identifies the 2nd cell in layer-4. Here i=1 indicates the layer nearest to and i=N indicates the layer farthest from the sink, whereas j=1 indicates the cell nearest to and j=M indicates the cell farthest from the sink within a layer. Bi is defined as the boundary between two consecutive layers- layer-i and layer-(i+1). For example, B1 (refer to Fig. 1) is the boundary between layer-1 and layer-2. Each boundary has a number of vertices e.g., in boundary Bi there are 2M such vertices. B. Assumptions Since the camera sensor nodes are more costly and heterogeneous deployment has been found as one of the efficient techniques for ensuring balanced energy consumption [7] therefore, in this work we assume deployment of static heterogeneous SNs to fulfill the said objective. The SNs are heterogeneous in terms of functionality. By functional heterogeneity, we mean a camera based visual sensing node (VN) captures the image from the environment, generates sensory data and periodically transmits those data to a neighbouring relay node (RN). A RN forwards the data received from both the neighbouring VNs and neighbouring RNs which are farther away from the sink. In this way sensory data from a VN reaches the sink through intermediate RNs. Further, we assume both a VN and a RN have initial energy as H0 while an unlimited amount of energy is set for the sink. We also consider always-on applications in which each VN uniformly generates n-bits packet and sends the packet to the sink through intermediate RNs at fixed time-interval. The RNs collect packet and send them to the sink through multi-hop communication after a time-interval of 1sec. Each VN and RN has the same transmission range rc and sensing range rs . Without loss of generality, during theoretical analysis an ideal MAC layer with no collisions and retransmissions is assumed for simplicity purpose. In our simulations, in addition to ideal MAC layer, a MAC protocol is considered for investigating the impact caused by it to make the assumption more realistic. C. Sensing and Communication Models We assume that a unit area is said to be covered, if every point in that area is within the viewing or sensing range of at least one active VN. The VNs perform observation at an angle of 3600. The maximal circular area covered around a VN is defined as its sensing area. The radius of the sensing area is called the VN’s sensing range rs . Further, we assume that the relationship between r and rs must satisfy the condition r d rs 3 for covering the network area (refer to Fig. 1) under consideration. We also assume that the network is considered as connected, if any active node (both VN and RN) can communicate with any other active node either in single-hop or in multi-hop. The relationship between r and rc must satisfy the condition r d rc 7 for ensuring connectivity in the network (refer to Fig. 1). Once the nodes (VNs/RNs) are deployed, network set-up phase starts. In this phase, all nodes learn about their one-hop neighbour RNs by exchanging messages among themselves. Here, one-hop neighbours RNs are those which are within the communication range of a node (VN/RN). Similar to the q-switch routing protocol [9], we consider that a VN in layer-

i communicates with a RN of the same layer for transmitting data via one hop and the sink via (i+1) hops. A VN in a layer chooses a RN as a forwardee (among the one-hop communication range) which is in the same layer and has the highest residual energy for sending its data. If there is more than one forwardee node with the same highest residual energy, one of them is chosen randomly. Next, the forwardee RN employs same procedure to choose the next forwardee RN in the next inner layer for sending its data. This process repeats till the data arrives at the sink. D. Energy Model The energy model specifies the energy consumption by a node during various operations such as radio transmission, reception, sensing and processing. Based on the observation for the experimental data and the nature of the circuit, we use the First Order Radio model [7] to estimate the energy consumption for sensing, processing, transmission and reception. Energy consumed for transmitting n-bits data to distance rc is e tx  n, rc n  eelec  eamp u rc2 n e t where e t e elec  eamp u rc2 is energy spent for transmitting one bit of data, eelec is the energy spent for activating the transmitter or receiver circuitry, eamp is the energy spent for the transmitter amplifier to communicate. During reception by a RN, energy consumption depends only on the data load, thus, energy consumed for receiving n-bits data is e rx  n n eelec n e r , where e r eelec is energy required to receive one bit of data. In addition to transmitting and receiving energy consumptions, energy consumption for sensing n-bits data around a distance rc is esx  n n es , where es is energy required to sense one bit of data. On the contrary, energy consumed for processing n-bits data is e px  n n e p , where e p is energy required to process one bit of data. Since image compression is also considered as an integral part of VSNs, therefore, energy model described above needs slight modifications. According to the experiments on energy consumption in image compression algorithm [10], an image compression enabled VN can compress the captured images and reduce the size of packets workload forwarded to the sink. To finish the image compression task, a VN has to spend extra processing energy. If k t is image compression ratio, then, the energy model considering image compression algorithm in our work can be summarized as: and e tx  n, rc k t n e t , e rx  n k t n e r , e px  n k com n e p esx  n n es , where k com is amplified gain for extra image compression operation needed other than the basic image processing function and k com ! 1 [7]. E. Network Lifetime In presence of several existing state-of-the-art definitions of network lifetime [11], we define the network lifetime of a VSN to be the time period during which the network incessantly satisfies the application requirement. The application requirement can be affirmed in several ways. One easy way to express the requirement of an application is in terms of coverage and connectivity. Therefore, in our case network lifetime is determined as the time till the proportion of dead nodes exceeds a certain threshold, which may fail to provide any coverage and/or connectivity in a certain region. We consider a node as dead when the amount of energy of the node is less than a predetermined threshold value. The threshold value for a VN is when it is neither able to sense any event nor able to transmit any data whereas for a RN it is neither able to transmit nor able to receive any data.

III.

ANALYSIS ON NETWORK LIFETIME

This section presents an analysis on network lifetime with an objective to find out the parameter(s) which have significant influence on energy balancing so that network lifetime can be extended by controlling the parameter values. Definition 3.1 (Energy Balance). Balanced energy depletion means that all nodes in the network deplete their energy simultaneously. Alternatively, if a network is energy balanced, the lifetimes of both VNs and RNs are the same and identical. Let the energy consumption rate (ECR) of VNs and RNs in layer-i is ECiVN and ECiRN , respectively. Further, let the i i total number of VNs and RNs in layer-i is TVN and TRN , respectively. If uniform energy consumption of each node within the same layer is achievable, then balanced energy depletion is achieved if i H0 u TVN i ECVN

k H0 u TVN k ECVN

i H 0 u TRN i ECRN

k H 0 u TRN  i, k, i z k.  k ECRN

The VNs of all the layers spend their energy in sensing, processing and transmitting own sensory data, whereas the RNs of all layers spend their energy for receiving, processing and transmitting. During analysis we consider shortest path routing policy as described in section II.C ignoring the part of choosing a forwardee with highest residual energy. Let us i i consider that TVN number of VNs and TRN number of RNs are deployed in layer-i. Now, from operational perspective, total energy consumption by the VNs of layer-i per unit timeinterval is given as 

ECiVN

i n TVN  es  k com ep  k t et 

 

for i=1,2,…,N. The RNs in the last layer consume energy for receiving, processing and transmitting of data received from VNs of own layer and the RNs of remaining layer consume additional energy for relaying data received from RNs located farther away from the sink. So, the total energy consumption per unit time-interval by the RNs of layer-i is computed as 

N h º ­ ª i °n TVN  ¦h i1TRN ¼  kt er  kcom ep  kt et for i 1,2,!,  N 1   ECiRN ® ¬ N °¯ nTVN for i N  kt er  kcom ep  kt et

Theorem 1: In RHC architecture, uniform energy depletion of VN and RN is guaranteed if the numbers of RNs are deployed in following numbers



i TRN

N i h º ­ ªTVN c  ¦ h i 1 TRN ¼ 1 for i 1, 2,..., N  1 °¬  ° c2  ® N ° TVN c1 for i N ° ¯ c2

 

where c1 k t  e r  e t  k com e p and c2 es  k com ep  k t e t . Due to page limitation, the proof of Theorem 1 could not be incorporated. For uniform energy depletion, the numbers of RNs are to be determined according to the relation given in (3). From (3), eventually it is seen that the deployed number of RNs in a layer depends on both the number of deployed VNs in that

layer and number of RNs deployed in the layers farther away i i ! TVN i.e., TRN . Also, for balanced energy consumption, RNs consume more energy compared to VNs due to the facts that relay load is shared by the RNs. Further, as the RNs are considerably cheaper than VNs [7], we may afford to deploy more number of RNs compared to VNs to mitigate energy hole problem. Since all VNs perform the same data collection procedure in a time interval, VNs deplete their energy at the same time. From the definition of lifetime (refer section II.E), the lifetime of a VN is computed using (1) and it is given as LT VN

i u H0 TVN ECiVN

LTiVN



H0

n es  k com e p  k t e t





In order to share equal relay load so that energy consumption remains uniform amongst RNs and VNs located in different layers, the required number of RNs can be calculated from Theorem 1. After deploying the calculated number of RNs in different layers (using (3)), energy consumption rates for all RNs in each layer are the same. Therefore, the lifetime of RNs located in layer-i (using (2)) is given as LTiRN

i n ª«TVN  ¦h ¬

N

i TVN u H0  h º T k e k e k e    RN t r com p t t i 1 ¼»

IV.

LT VN 

DEPLOYMENT SCHEME

In the proposed deployment strategy, nodes are deployed in three phases. In first phase, VNs are deployed in some strategic locations which are located on the boundary of each layer for ensuring 1-coverage of the network. This deployment ensures minimum number of VNs for covering the entire network area. In second phase, RNs are placed at the centre of each cell of each layer for ensuring connectivity of the network. The remaining RNs are deployed in third phase which are uniform randomly distributed within each cell area. The prime objective of the proposed node deployment scheme is mitigating energy hole problem by achieving balanced energy consumption of VNs and RNs so that network lifetime is enhanced. A. Location of Node Deployment As mentioned earlier, the sink is assumed to be located at one corner of the network area with co-ordinate (0, 0). The location of VNs and RNs are calculated with respect to the sink’s location. The location (x, y) where a VN or a RN is to be placed can be computed by controlling the different parameters in the following equations:  

x

y

Cn

1· § 3 r u ¨ j  ¸  Cn  2¹ ©

 

r ª 2  Q  3  i  1 º  ¼ 2¬

 

where i is the layer number, j is the cell number for each layer-i and r is the cell radius.

­SVN  ® ¯SRN

In (4), C n is used to choose whether VNs or RNs will be deployed. 

­° 3 r ® 2 °¯ 0

SVN

for even layer  for odd layer



Here, SVN will be used for determining the locations of VNs at strategic locations in a layer. Further, the value of SVN decides whether the VNs are to be deployed in odd or even layer. For example, SVN is 0 for i=1 and it is  23 r for i=2. 

­° 3 r ® 2 °¯ 0

SRN

for odd layer  for even layer



Similar to SVN , the value of SRN will be used for determining the locations of RNs at the center of each cell in a layer. Based on even or odd layer, SRN also has two values e.g.,  23 r for i=1 and it is 0 for i=2. 

Since all the VNs and RNs in the network consume energy uniformly thus, in our work, the network lifetime is expressed as LTiRN

The other variables used in the above expressions (refer (4) and (5)) are as follows:

Q

­3 ® ¯2

for VNs for RNs





Here, Q signifies whether VNs or RNs are to be deployed. For example, at the strategic location on the boundary as VNs are deployed, the value of Q is 3. Similarly at the centre of each cell as RNs are deployed, the value of Q is 2. After deploying the RNs at the centre of each cell, if excess RNs remain, those are deployed uniform randomly within the cell area using the following equations

x rand

xc 

3r rand(0, 1) u cos(2S rand(0, 1))  2

 



y rand

yc 

3r rand(0, 1) u sin(2S ran (0, 1))  2

 

where x c and yc are the co-ordinates of the centre of a cell, whereas the locations of the centre of a cell can be found by using (4) and (5). B. Quantification of Nodes The objective of our present node deployment is to maximize network lifetime by maintaining balanced energy consumption throughout the network. In this section, we have derived the total numbers of VNs and RNs required in the network for balanced energy consumption amongst VNs and RNs. Theorem 2: For a given N-layer RHC network where each layer divided into M number of cells, the total number of VNs and RNs required for balancing energy consumption while maintaining coverage and connectivity is  M u N and ª ª§ « M «¨ c1 « «© c 2 « ¬

· § c1 · N  N  1 § c1 · ¨ ¸ ¸N¨ ¸ 2 ¹ © c2 ¹ © c2 ¹ 2

3

 N  1  N  2  "  § c1 · N º» º» , 2

¨ ¸ © c 2 ¹ »¼ »»

respectively. Due to page limitation, the proof of Theorem 2 could not be incorporated.

V.

PERFORMANCE EVALUATION

Performance of the present Location-wise Predetermined Heterogeneous Node Deployment (LPHND) strategy, reported in the earlier section is measured based on two parameters- energy balancing and network lifetime. A. Simulation Arrangement For performing the experiments Prowler-Rmase [12], [13] network simulator has been used. The transmission range rc is taken as 20m, initial energy in each sensor H0 is 10J. We conducted the experiments considering two network sizes- one consisting of 4 layers with each layer having 7 RHCs and the other consisting of 7 layers with each layer having 5 RHCs. Therefore, from Theorem 2, we deployed 28 and 35 VNs whereas 116 and 881 RNs, respectively, for these two network sizes. Remaining parameters used during simulation are shown in Table I. We focus on the following metrics: a) Avg ECR per VN and Avg ECR per RN in a layer defined as the average energy spent by each node (VN/RN) in a layer during the network operation; b) network lifetime as defined in section II.E. First metric is useful for determining whether the energy consumption in a network is balanced and second metric is for consequent enhancement of network lifetime. Two sets of experiments are conducted for evaluating the performance of the proposed deployment strategy. One set of experiment measures energy balancing in the network and the last set verifies the enhancement of network lifetime. Simulation results of our strategy LPHND are compared with an existing node deployment strategy, namely, Relay Node Deployment with Gaussian Distribution (RNDGD) [7]. In order to have an integer number of RNs for each layer, upper ceil function is employed in (3). Extensive simulation has been performed with a confidence level of 95% and average of 200 independent runs has been taken while plotting the simulation graphs.

Deployed RNs in each layer

B. Analyical Result In this section, we compared LPHND with competing scheme RNDGD to observe at what extent our scheme achieves the design goal of deploying RNs in the network to mitigate energy hole problem. However, it is assumed that in both the schemes, minimum numbers of VNs are uniform randomly deployed in the network area ensuring 1-coverage. 450 400

LPHND

350

RNDGD

300 250 200 150 100 50 0 1

2

3

4

5

Layer number

6

7

Fig. 2. RNs deployed in each of the layer when RNs are deployed using LPHND and RNDGD for 7-layer network.

Figure 2 is the analytical result of the number of RNs deployed in each layer for LPHND and RNDGD. Our

TABLE I PARAMETER VALUES USED IN SIMULATION Parameters

Values

Transmitting energy per bit  e t

60nJ

Receiving energy per bit  er

50nJ

Processing energy per bit  ep

50nJ

Sensing energy per bit  es

50nJ

Image compression energy factor  k com

1.2

Image compression ratio  k t

55%

Radius of each RHC  r Gaussian standard deviation (σ) Packet size (n) Network size (a×b)

7.56m 200 400-bits (85×50)m2 ~ (55×85)m2

primary observation is that, except the scheme RNDGD, in LPHND, the number of deployed RNs reduces in layers as the distance of layers from the sink increases fulfilling the objective of deploying more RNs near the sink. It is also observed that RNDGD gives more redundant RNs from layer-2 to layer-6. Furthermore, when RNs are deployed using RNDGD, it fails to deploy more redundant RNs in layer-1 which is very essential to combat the energy hole problem and enhance network lifetime. C. Simulation Results and Discussion While simulating LPHND and RNDGD, we considered both ideal and realistic scenarios. Here, by ideal scenario we mean the routing protocol considered in section III and without any MAC protocol. On the flip side, by realistic scenario we mean routing protocol as described in section II.C and funneling-MAC [14] as a MAC protocol. The funneling-MAC is a hybrid MAC protocol where TDMA (schedule-based) is used in nodes located within a few hops from the sink whereas CSMA/CA (contention-based) is used in nodes located far away from the sink. We consider nodes located within layer-2 use TDMA whereas nodes beyond layer-2 use CSMA/CA. Furthermore, we have considered energy consumption rate as 5% and 2.5% of the energy consumption rate of reception for idle and sleep mode respectively. In simulation plots, a scheme name with ‘(R)’ signifies performance under realistic scenario and without it under ideal scenario. D. Energy Balancing Figure 3 shows avg ECR per VN for 4-layer network. As expected, the gradient of the curve for avg ECR per VN is flat throughout the network. Further, the overlapping nature of the plots demonstrates that irrespective of the schemes, VNs consume energy uniformly. It is due to the fact that in both LPHND and RNDGD schemes VNs are deployed uniform randomly throughout the network area and even if the network size is varied, the gradient of the curve for avg ECR per VN remains flat. On the contrary, from Fig. 4 it is observed that in LPHND for both ideal and realistic scenarios, the avg ECR per RN of a particular network size is constant for all layers and this rate varies with network sizes. Precisely, avg ECR per RN increases with increase in network size. For example, in case of ideal scenario, avg ECR per RN is 42.17μJ/sec for 4-layer network area whereas for 7-layer network it is 44.2μJ/sec. Similarly, in realistic scenario, avg ECR per RN is 51.4μJ/sec for 4-layer network

whereas for 7-layer network it is 53.1μJ/sec. In RNDGD it is observed that avg ECR per RN varies in different layers for a given network size. Furthermore, in RNDGD, irrespective of network size, RNs in layer-1 have maximum avg ECR and RNs in farthest layer have the lowest avg ECR. This justifies our claim that LPHND is relatively more energy balanced compared to the competing scheme RNDGD. Avg ECR per VN (μJ/sec)

85 80 75 70

LPHND

65

LPHND (R)

RNDGD RNDGD (R)

60 55 1

2

3

4

Layer number

Fig. 3. Average energy consumption rate per VN under ideal and realistic scenarios for 4-layer network. Avg ECR per RN (μJ/sec)

Avg ECR per RN (μJ/sec)

425

LPHND RNDGD LPHND (R) RNDGD (R)

375 325 275 225 175 125 75 25 1

2

3

Layer number

450 400 350 300 250 200 150 100 50 0

REFERENCES

RNDGD LP HND (R)

[1]

RNDGD (R)

[2]

[3] 2

3

4

5

Layer number

6

7

(a) (b) Fig. 4. Average energy consumption rate per RN under ideal and realistic scenarios for various network sizes. (a) 4-layer network. (b) 7-layer network.

[4]

Network lifetim e (hours)

Network lifetim e (hours)

[5] 140 LP HND R NDGD LP HND (R ) R NDGD (R)

120 100 80 60 40 20 0 1

2

3

Layer number

4

350 LP HND RNDGD LP HND (R) RNDGD (R )

300 250 200

[6]

150 100

[7]

50 0 1

2

3

4

5

Layer number

6

CONCLUSION AND FUTURE WORK

In this paper, we have proposed a location-wise predetermined heterogeneous SNs deployment scheme in VSNs. The target of the scheme is to alleviate the energy hole problem by balancing energy consumption among SNs and subsequent enhancement of network lifetime while maintaining coverage and connectivity. We have analyzed the issue of enhancement of network lifetime by limiting the energy hole problem. Analysis revealed number of realy nodes and their location has significant influence on limiting the energy hole problem and enhancement of network lifetime. Considering the results of this analysis, we have developed a location-wise predetermined deployment scheme using heterogeneous nodes. Simulation results prove that the proposed strategy outperforms an existing Gaussian distribution based relay node deployment scheme [7] with respect to energy balancing in the network. As a future extension of our work, the deployment strategy may be made more realistic by considering node placement error. Also, the scheme may be analyzed further for improvement considering performance metrics e.g., end-to-end delay, packet loss and throughput.

LP HND

1

4

VI.

7

[8]

(a) (b) Fig. 5. Network lifetime under ideal and realistic scenarios for various network sizes. (a) 4-layer network. (b) 7-layer network. [9]

E. Network Lifetime The graphs illustrated in Fig. 5 represent the network lifetime for two different network sizes. It is observed irrespective of the scenarios, that with increase in network size, network lifetime decreases e.g., in realistic scenario for 4-layer network, it is 54.14 hours whereas for 7-layer network it is 52.21 hours. This is due to the fact that with increase in network size, the nodes in the innermost layer need to relay increased volume of data from the outer layers thereby causing more energy consumption. Moreover, in LPHND the flat nature of the plot ensures that in all the layers, network lifetime terminates in more or less the same time as compared to RNDGD. This ensures that energy in LPHND is balanced to a greater extent.

[10]

[11]

[12] [13] [14]

S. Soro and W. Heinzelman, “A Survey of Visual Sensor Networks,” Advances in Multimedia, vol. 2009, no. 21, May 2009. I. F. Akyildiz, T. Melodia and K. R. Chowdhury, “A Survey on Wireless Multimedia Sensor Networks,” Computer Networks, vol. 51, no. 4, pp. 921–960, Mar. 2007 S. Halder, A. Ghosal and S. DasBit, “A Pre-determined Node Deployment Strategy to Prolong Network Lifetime in Wireless Sensor Network,” Computer Communications, vol. 34, no. 11, pp. 12941306, Jul. 2011. Y. Charfi, N. Wakamiya and M. Murata, “Challenging Issues in Visual Sensor Networks,” IEEE Wireless Communications, vol. 16, no. 2, pp. 44 –49, Apr. 2009. S. Soro and W. Heinzelman, “Camera Selection in Visual Sensor Networks,” Proc. of Int’l Conf. on Advanced Video and Signal Based Surveillance, pp. 81-86, Sep. 2007. E. Yildiz, K. Akkaya, E. Sisikoglu and Ismail Guneydas, “Camera Deployment for Video Panorama Generation in Wireless Visual Sensor Networks,” Proc. of Int’l Symposium on Multimedia, pp. 595600, Dec. 2011. H. Li, V. Pandit and D. P. Agrawal, “Gaussian Distributed Deployment of Relay Nodes for Wireless Visual Sensor Networks,” Proc. of IEEE Global Communications Conference, pp. 5374-5379, Dec. 2012. M. Hooshmand, S. M. R. Soroushmehr, S. Samavi, P,. Khadivi and S. Shirani, “Visual Sensor Network Lifetime Maximization by Prioritized Scheduling of Nodes,” J. of Network and Computer Applications, vol. 36, no. 1, pp. 409-419, Jan. 2013. X. Wu, G. Chen and S. K. Das, “Avoiding Energy Holes in Wireless Sensor Networks with Nonuniform Node Distribution,” IEEE Trans. on Parallel and Distributed Systems, vol. 19, no. 5, pp. 710-720, May 2008. L. Ferrigno, S. Marano, V. Paciello and A. Pietrosanto, “Balancing Computational and Transmission Power Consumption in Wireless Image Sensor Networks,” Proc. of Int’l Conf. on Virtual Environments, Human-Computer Interfaces and Measurement Systems, pp. 61–66, Jul. 2005. I. Dietrich and F. Dressler, “On the Lifetime of Wireless Sensor Networks,” ACM Trans. on Sensor Networks, vol. 5, no.1, pp. 1-39, Feb. 2009. Prowler,[Online].http://www.isis.vanderbilt.edu/projects/nest/prowler. Rmase,[Online].http://www2.parc.com/isl/groups/era/nest/Rmase. G. S. Ahn, E. Miluzzo, A. T. Campbell, S. G. Hong and F. Cuomo, “Funneling-MAC: A Localized, Sink-Oriented MAC For Boosting Fidelity in Sensor Networks,” Proc. of ACM Int’l Conf. on Embedded Network and Sensor System, pp. 293-306, 2006.