MOBILE SENSORGROUP WITH SMART PATH FOR

DETECTING TARGET AREA

Kuen-Liang Sue and Jing-Wei Lin

Department of Information Management, National Central University, Taoyuan, Taiwan

Keywords: Mobile Sensor Networks, Incremental Clustering, Exploration.

Abstract: The aim of this study is to design an intelligent mechanism for exploring unknown target areas in a plant,

such as oil pollution in ocean. To explore the target area efficiently, a smart sensing mechanism based on

Incremental Clustering algorithm is proposed to cooperate with a small sensing network structure named

Centralized SensorGroup (CSG). The operation processes and detection phases are provided and verified in

the investigation. System performance is evaluated by observing the detection completeness and accuracy in

different scenarios within a square experimental area of 100m*100m. No matter when large or small

scenarios are explored, simulation results demonstrate that CSG cooperated with the smart sensing

mechanism has quite good detection accuracy and efficiency and can achieve the purpose of exploring

target area efficiently and effectively.

1 INTRODUCTION

With the development of robot and wireless

transmission technologies, mobile sensors (MS) can

provide mobility for wireless sensor networks and

mobile sensor networks can be formed through

various wireless interfaces

(Liang et al, 2006; Dantu

et al, 2005

; Hu and Evans, 2004). Mobility of

mobile sensor networks brings some significances

and potentials. Mobility can let the sensor network

moves to collect information in the environments

that traditional fixed sensor networks can not be

deployed (Clark, 2005; Casbee et al, 2006; Wang et

al, 2005). One of the application issues of mobile

sensor network is to detect and localize specific

target areas in the sensing environment, for example,

the detection of oil spoiled area in the ocean

environment. Traditional sensor networks detect

environment by random deployment. However, due

to the higher cost of MS, it is inappropriate to use

mobile sensor network by the same way. Hence,

how to use few MS nodes to detect unknown target

areas efficiently is a key issue in this application.

The aim of this study is to design an intelligent

mechanism for exploring unknown target areas in a

plant, such as oil pollution in ocean. To detect the

targets efficiently, a smart sensing mechanism based

on Incremental Clustering algorithm is proposed to

cooperate with a small sensing network structure

named Centralized SensorGroup (CSG).

The remainder of this paper is structured as

follows. Because we use incremental clustering

algorithm to improve the efficiency of detection, the

related data clustering techniques is introduced

briefly in section 2. The proposed mobile sensor

network structure and sensing mechanism are

described in section 3. Simulation method and

numerical analyses are described in section 4.

Section 5 provides conclusion for the investigation.

2 BACKGROUNDS

In this study, we propose a sensing mechanism

designed for a small mobile sensor network to detect

the unknown target areas in the environment. The

proposed mechanism utilizes a skill for data mining

called cluster analysis to offer sensor network the

ability to analyze the distribution of target area from

sensing records and adjust the movement of CSG

dynamically by using data clustering algorithms.

Hence, the exploration job can be performed more

efficiently and effectively.

Cluster analysis is an unsupervised machine

learning technology. Clustering algorithms’

operation is based on the concept of grouping the

similar data into a cluster and making the data in

different clusters dissimilar (Jain et al, 1999). The

Sue K. and Lin J. (2010).

MOBILE SENSORGROUP WITH SMART PATH FOR DETECTING TARGET AREA.

In Proceedings of the International Conference on Wireless Information Networks and Systems, pages 36-41

DOI: 10.5220/0002981100360041

 SciTePress

traditional techniques are mainly designed for static

databases environment. For example, the K-means

algorithm is one of the traditional partitioning

clustering algorithms (Law and Jain, 2005). The

general objective is to obtain a fixed number of

clusters and minimizes the total square errors of the

all clusters.

Figure 1: CSG structure.

However, with the increasing requirements of

data clustering, the limits of traditional data

clustering algorithms exist due to that they usually

need to load all data into memory while analyzing.

To solve the problem, incremental clustering

algorithms are proposed for the dynamic database

environments (Ester et al, 1998; Pons-Porrata et al,

2005). Incremental clustering algorithms executes

clustering analysis whenever a data enters the

database, thus these algorithms can cluster all data

by executing simple calculations instead of loading

all data into memory and executing complicate

analysis. That is, incremental clustering algorithm

needs fewer requirements for computation resource

and ability. The proposed exploration mechanism is

supposed to cooperate with mobile sensors which

are usually assumed to have limited computing

resources. Therefore, incremental clustering

algorithm is appropriate to be used for activating the

sensing mechanism in the proposed sensing

networks.

3 EXPLORATION SCHEME

To detect the targets efficiently, the mobile sensors

should be formatted to a network and operate as a

group basing on a systematic sensing mechanism.

Thus this study is composed with two parts - mobile

sensor network structure and sensing mechanism.

3.1 Centralized SensorGroup

Centralized SensorGroup (CSG) is a mobile sensor

network structure constructed with number of MS

nodes crowded within a specific range. The concept

of CSG is to let CSG keeps moving in the

environment. MS nodes in a CSG keep detecting

targets while moving. If a MS detects the border of

target, then it records its current location as a

sensing record. Finally, the location of targets’

borders can be known by using the sensing records.

Figure 2: Initial phase.

In the following parts, we introduce the CSG

structure and CSG’s operation process.

3.1.1 CSG Network Structure

The movement of MS in CSG is based on Reference

Point Group Mobility (RPGM) (Hong et al, 1999).

According to the functions of MS in CSG, there are

two roles of MS in CSG called Center node (C node)

and Member node (M node). As shown in Fig. 1, C

node is the MS allocated in the center of CSG. It is

responsible of controlling all operations of CSG and

broadcasting its location to all M nodes within its

transmission range (TR) periodically. M nodes are

the other MS deployed surround C node. They move

randomly surround C node within the range of group

range (GR). In CSG structure, only C node requires

GPS ability. C node acts as a mobile anchor in the

network and other M nodes can calculate their

locations by using a wireless localization algorithm.

We assume that all MS in CSG should be equipped

with a compass to measure the direction of its

movement and have the ability to measure the

distance of its movement.

3.1.2 Localization Algorithm of CSG

The localization algorithm used in CSG is modified

from the algorithm proposed (Akcan et al, 2006). In

this algorithm, all mobile nodes within a group

exchange their information with each other and

calculate their local coordinates in the group basing

MOBILE SENSORGROUP WITH SMART PATH FOR DETECTING TARGET AREA

on the information to keep their group moving. In

this study, we modify this algorithm and combine it

with CSG structure. There are two phases

in the modified algorithm, the “initial phase” and the

“verification phase”.

In the initial phase, C node broadcasts its

location with packets Loc1 and Loc2 at time t1 and

t2 sequentially. As shown in Fig. 2, C node adds its

Figure 3: Verification phase.

location (x0, y0) into Loc1 and broadcasts it to all M

nodes at time t1. An M node can measure the

distance to C node d1 by TOA (Time Of Arrival)

when it receives Loc1. After t1, C node and M nodes

start moving along the directions they decided. At

time t2, C node then broadcast its location (x1, y1)

with Loc2 again and an M node measures the

distance to C node d2 when it receive Loc2. After

the above process is finished, an M node can

calculate two circles C1 and C2 by using the

coordinates (x0, y0), (x1, y1) and the distance d1,

d2. These two circles define the sets of possible

locations of an M node at t1 and t2. Because M node

can calculates it displacement vector from t1 to t2, it

can calculates a new circle C1’ by displacing C1

with the vector (v2cosβ, v2sinβ). Because both C1’

and C2 define the set of possible locations of M

node at t2, M node can get its t2 candidate locations

by calculating the intersection of C1’ and C2.

In the verification phase, each M node exchanges

its t2 candidate locations with each other by

broadcasting a VeriInfo packet. When an M node

receives VeriInfo from other one, it firstly measures

the distance to the sender, then verifies an answer

from its candidate locations by using this distance.

Finally, M node calculates a weight for the verified

answer. For example, as shown in Fig. 3 (a), M

receives a VeriInfo packet from M’. M firstly

measures the actual distance d3, and then uses d3 to

verify the most possible candidate. As shown in Fig.

3 (b), M can compose at most four sets of candidate

location by using the t2 candidate locations of M and

M’. Then M compares the distances of four

candidate sets a, b, c and d with d3 and chooses the

set which has smallest difference with d3 as the

verified answer. Then M gives this answer a weight

value by calculating the inverse of the difference of

the chosen candidate set distance and the actual

distance d3. As shown in Fig. 3 (b), the candidate set

with the distance value “a” has the smallest

difference with d3, so M chooses (x3', y3') as the

verified answer of the VeriInfo packet from M’ and

gives (x3', y3') a weight by calculating 1/|d3– a|. An

Figure 4: CSG operation procedure.

M node can choose one of its candidates and gives

the one a weight value whenever it receives a

VeriInfo. The weight value for each candidate

location is accumulated. Finally, M can get the t2

localization result by calculating the weighted mean

of its candidate locations and can also get t1

localization result by using its displacement vector

from t1 to t2 in the initial phase.

3.1.3 CSG Operation Process

The CSG operation procedure and communi-cation

protocol between CSG members are shown in Fig. 4.

In the initial of operation, CSG is firstly set to the

initial location (step 1). Then C node starts moving

with speed Vc and leads CSG to detect targets (step

2). CSG iterates step 4-14 with a period of T seconds,

this periodical process is called a “RPGM round”. In

this process, C firstly gets its GPS location Lc and

broadcast Lc with a RPGMbroadcast packet to all M

nodes (step 4-5). When an M node receives a

RPGMbroadcast (step 7), it firstly executes the

process to end the n-1th RPGM round. This process

includes the localization result determination of n-

1th RPGM round (step 8) and the conversion of the

temporary sensing records in n-1th RPGM round

(step 9). In step 9, due to that M nodes in CSG can’t

get their current locations timely, all sensing record

in n-1th RPGM round are temporarily stored as the

displacement vectors from time t

to the time they

detect the border of target. After the t1localization

(

)

(

)

Centralized SensorGroup procedure

1. Set initial location of CSG

2. C moves with speed V

3. iterate every T seconds period

4. L

Å GPS location of C

5. C RPGMbroadcast ( L

)

6. for each M do

7. Receives L

form C

8. Location determination

9. Records conversion

10. L

m_init

Å initial location

11. L

m_dest

ÅRandom destination

12. M moves with speed V

13. CSG Localization process

WINSYS 2010 - International Conference on Wireless Information Networks and Systems

result is determined in step 8, M nodes convert the

temporary sensing records in n-1th RPGM round to

actual sensing record by adding t

localization result

to all temporary records. M node then starts the

process of nth RPGM round. It calculates the initial

location L

init

of nth RPGM round by adding t

localization result of n-1th RPGM round with the

displacement form t

to the time it receive

RPGMbroadcast of nth RPGM round (step 10).

Figure 5: Concept of data collection phase.

Then M node decides a random destination L

dest

around L

within range of GR (step 11) and starts its

movement with speed V

(step 12). CSG then

executes the localization process of nth RPGM

round (step 13), C node sends Loc1 after

RPGMbroadcast immediately at t

. After T seconds,

C node broadcasts RPGMbroadcast again to end the

nth and start the n+1th RPGM round.

3.2 Sensing Mechanism

To detect the unknown targets in the environment, C

node need a systematic sensing mechanism to

control the moving path of CSG. An Incremental

Clustering Aided Sensing Mechanism (ICASM) is

proposed in this study. In this mechanism, we

assume that there is no obstacle in the environment.

There are two phases in a round of ICASM’s

operation. The first phase is “data collection phase”.

CSG executes a quick scan of the whole

environment to analyze the rough locations of the

targets’ borders. The second phase is “Detailed

detection phase”. CSG make a detailed detection

along the rough location of targets’ borders

according to the analysis of the first phase.

In the data collection phase, C node splits the

whole square sensing environment into number of

rectangle regions with the width of GR*2. The

centerlines of the regions are the C node’s moving

paths in this phase. C node leads CSG to scan the

environment along these paths. Because it is

possible that CSG can’t detect the borders of each

region effectively, C node splits the regions with an

interlaced style in odd and even round of ICASM.

According to the split regions, the border of

unknown target can also be split into number of

deformed lines. We expect that all sensing records

collected in this phase should be clustered basing on

these lines, each lines can own one or more clusters.

Finally, as shown in Fig. 5, C node can catch the

rough location of the target’s border by using the

(a) Single large area (b) Discrete small area

Figure 6: Scenario setting.

centroid of each cluster. To reduce the load of C

node to cluster records, a simple incremental

clustering algorithm is used as the clustering method

in this mechanism.

4 SIMULATIONS

4.1 Scenario Settings

The environment in the simulation is a 100 m*100 m

square area. There are two types of target area

deployed in the sensing environment. The first is

“single large target area”, as shown in Fig. 6 (a), the

coordinates are the apexes of target. The other is

“discrete small target area”, as shown in Fig. 6 (b),

each target is a rhombus with width and height of 2

m and the coordinates are the centroid of each target.

4.2 Parameter Settings

We use NS2 as the simulation platform. The

simulation time is 6000 seconds. A CSG network is

composed of one C node and three M nodes; the

transmission range of a node is 15 m, the period of a

RPGM round is 1 second, the speed of C node and

M nodes are 3 m/s and 5 m/s and the range of CSG

is 2 m. In the localization algorithm of CSG, the

execution time of initial phase and verification phase

in a RPGM round are 0.9 and 0.1 second.

In the localization algorithm of CSG, each MS

MOBILE SENSORGROUP WITH SMART PATH FOR DETECTING TARGET AREA

uses wireless signals and compass to measure the

distance between each them and their moving

directions. Both of them may be influenced by

interferences in environment. To simulate the

interferences from environment, random distance

error and angle error are added into the localization

process to analyze the effect of different level of

interferences. The range of distance error is -10% to

10% and range of angle error are -10 to 10 degrees.

Accuracy of GPS is assumed always accurate.

Figure 7: Detection accuracy.

4.3 Simulation Analysis Methods

We use “detection accuracy” and “detection

completeness” to analyze the detection performance

of CSG. The detection accuracy is evaluated by

calculating the average error of all sensing records

collected by CSG. We define the error of a sensing

record as the shortest distance for the record to the

border of target area.

To evaluate the of detection completeness are set

anchor coordinates on the border of target per 0.2m.

If one or more sensing record exists around an

anchor within 0.2 m, we set the status of the anchor

to “sampled”. Hence, we evaluate the detection

completeness by calculating the ratio of the number

of sampled anchor and the total number of anchor

after matching the anchor coordinates with all

sensing records.

4.4 Simulation Results

4.4.1 Detection Accuracy

Because the sensing records are mainly collected by

the M nodes, the detection accuracy is decided by

the localization accuracy of M nodes. The average

detection accuracy of CSG structures using ICASM

in two target area scenarios is shown in Fig. 7. We

can see that the increasing localization error of M

nodes in CSG lifts the error of sensing records and

decreases the detection accuracy. The average error

of sensing records in single large target scenario and

discrete small targets scenario are 0.52 m and 0.48 m

when the distance error and angle error are set to

10% and 10 degrees. In both scenarios, the higher

level of interferences is, the lower detection

accuracy in CSG becomes.

(a) Detection completeness with time

(b) Detection result for 80% completeness

Figure 8: Detection completeness in large scenario.

4.4.2 Detection Completeness

In this part, we evaluate the relationship of detection

completeness and detection time of CSG using

ICASM in both two scenarios without interferences

(0 distance error and 0 angle error). Fig. 8 (a) is the

relationship of detection completeness and detection

time in single larger target scenario. The average

time of a round in ICASM is 1195 seconds and the

average detection time required to achieve 80%

detection completeness is 2258 seconds. During the

first round, because ICASM only detects the targets

roughly in the first phase, so the trend of detection

completeness increases slowly. However, in the

second phase, ICASM can let CSG to make a

detailed detection according to the analysis of first

phase. Thus the detection performance of ICASM

increases significantly in this phase. Beside, the

trend of detection completeness in ICASM can still

increase efficiently after the first round. This means

that ICASM can adjust the movement of CSG and

achieve the purpose of efficient detection. Fig. 8 (b)

is a 80% detection completeness graphic result of

sensing records collected by CSG in once simulation.

The detection time is 2234 seconds.

WINSYS 2010 - International Conference on Wireless Information Networks and Systems

Fig. 9 (a) is the relationship of detection

completeness and detection time in discrete small

targets scenario. The average time of a round in

ICASM is 1139 seconds and the average detection

time required to achieve 90% detection

(a) Detection completeness with time

(b) Detection result for 90% completeness

Figure 9: Detection completeness in small scenariois 2353

seconds. Fig. 9 (b) is a 90% detection completeness

graphic result of sensing records collected by CSG in once

simulation. The detection time is 2215 seconds.

5 CONCLUSIONS

To identify target areas automatically is an important

application of mobile sensor networks, especially in

abnormal environment such as oil pollution in

oceans. The research investigates the problem by

utilizing small mobile sensor network. Centralized

SensorGroup (CSG) organized with several sensor

nodes, only one of them called center node needs to

be equipped with GPS functionality. During the

exploring process, GPS-free member nodes can

localize themselves by computing information from

center node and their historic data. Hence, the sensor

network is rather cost-effective. Furthermore, a

sensing mechanism based on the incremental

clustering algorithm is also proposed to adjust the

moving direction of CSG dynamically according to

the distribution of the target area. The proposed

sensing mechanism can achieve the target area

exploration more efficiently.

The detection performance is evaluated by

detection completeness and accuracy for different

scenarios in a 100m*100m square environment. The

simulation results show that. The average detection

times to achieve 80% detection completeness are

2258 seconds for large polygon scenarios. In

discrete small target scenarios, the average detection

times to achieve 90% is 2353 seconds. The average

record errors which represent the detection accuracy

in different scenarios vary between 0.48m and

0.52m. To sum up, simulation results demonstrate

that CSG cooperated the ICASM has quite good

detection accuracy and efficiency.

ACKNOWLEDGEMENTS

The research was supported by the National Science

Council, Taiwan, under the contract NSC 98-2410-

H-008-004.

REFERENCES

Liang, Y. et al. (2006). A Review of Control and

Localization for Mobile Sensor Networks.

Proceedings of Intelligent Control and Automation, 2,

9164-9168.

Dantu, K. et al. (2005). Robomote: enabling mobility in

sensor networks. Proceedings of Information

Processing in Sensor Networks, 404-409.

Hu, L. and Evans, D. (2004). Localization for mobile

sensor networks. Proceedings of the 10th conference

on Mobile computing and networking, 45-57.

Clark, J. and Fierro, R. (2005). Cooperative hybrid control

of robotic sensors for perimeter detection and tracking.

2005 American Control Conference, 5, 3500–3505.

Casbee, D. W. et al. (2006). Cooperative forest fire

surveillance using a team of small unmanned air

vehicles. International Journal of Systems Sciences,

37(6), 351-360.

Wang, G. et al. (2005). Sensor relocation in mobile sensor

networks. Proceedings of the 24th IEEE Computer

and Communications Societies, 4, 2302-2312.

Jain, A. K. et al. (1999). Data Clustering: A Review. ACM

Computing Surveys, 31(3) 122-130.

Law, H. C., Jain, A. K. 2005. Data Clustering: A User’s

Dilemma. Proceedings of Pattern Recognition and

Machine Intelligence, 3776, 1-10.

Ester, M. et al. (1998). Incremental Clustering for Mining

in a Data Warehousing Environment. Proceedings of

Very Large Databases, 24-27.

Pons-Porrata, A. et al. (2005). An Incremental Clustering

Algorithm Based on Compact Sets with Radius α.

Proceedings of Progress in Pattern Recognition,

Image Analysis and Applications, 3773, 518-527.

Hong, X. et al. (1999). A group mobility model for ad hoc

wireless networks. Proceedings of the 2nd

international workshop on Modeling, Analysis and

simulation of wireless and mobile systems, 53–60.

Akcan, H. et al. (2006). GPS-Free node localization in

mobile wireless sensor networks. Proceedings of the

5th international workshop on Data engineering for

wireless and mobile access, 35-42.

MOBILE SENSORGROUP WITH SMART PATH FOR DETECTING TARGET AREA