A Dynamic Distributed Tree Based Tracking
Algorithm for Wireless Sensor Networks
1
Aysegul Alaybeyoglu,
1
Aylin Kantarci,
2
Kayhan Erciyes
1
2
Ege University, Computer Eng. Dept., Bornova, Izmir, Turkey
email: {aysegul.alaybeyoglu, aylin.kantarci}@ege.edu.tr
Izmir University, Computer Eng. Dept., Uckuyular, Izmir, Turkey
email: [email protected]
Abstract. We propose a dynamic, distributed tree based tracking algorithm for very fast moving targets in wireless sensor networks, with
speeds much higher than reported in literature. The aim of our algorithm is to decrease the miss ratio and the energy consumption while
tracking objects that move in high speeds. In order to do this, the root
node which is determined dynamically in accordance with the node’s distance to the target, forms lookahead spanning trees along the predicted
direction of the target. As the miss ratio decreases, the usage of recovery
mechanisms which are employed to detect a target again that is moving
away from the predicted trajectory also decreases. This decrease reduces
the energy consumption and increases the network lifetime. We describe
all the phases of the algorithm in detail and show by simulations that
the proposed algorithm performs well to track very fast moving targets.
We also compare the algorithm with the generic cluster, generic tree and
dynamic multi cluster based tracking algorithms in terms of miss ratio
and energy consumption.
Key words: target tracking, wireless sensor networks,localization
1
Introduction
Recent technological advancements made sensor nodes cheap and readily available for academic and industrial usage. Wireless Sensor Networks may consist of
thousands of nodes deployed in a large area. Sensor nodes are suitable for various
application types due to their sensing and wireless communication capabilities.
Military surveillance, habitat monitoring and target tracking are some of the
important types of applications for sensor networks [1, 2].
In target tracking applications, when a mobile target is sensed by some of the
nodes, its position is calculated by cooperation of these nodes using localization
techniques and aggregated data is sent to the sink node. There are many kinds
of tracking algorithms in literature[3-6]. Tree based tracking is one of the type
of these algorithms that underlies the proposed tracking algorithm.
In the generic spanning tree based tracking approach [7], spanning trees are
formed dynamically as the events occur in the network area. A root node is
2
1
Aysegul Alaybeyoglu,
1
Aylin Kantarci,
2
Kayhan Erciyes
selected from the nodes that detect the target. The root collects the sensed data
and calculates the location, speed and the direction of the target. With these
information, it predicts the next location and sends a warning message to the
node closest to the target’s predicted location. The node receiving this warning
message, becomes the new root node and forms its spanning tree to be ready to
detect the target.
In the existing studies, the current root node at time t predicts the location
only for time t+1. In case the target moves at high speed, it can pass by a group
of nodes very fast without being detected. Therefore in these studies, as the
target increases its speed, the probability of missing that target also increases.
Due to the high probability of target miss, existing tracking studies uses recovery
mechanisms frequently for target missing conditions. Using recovery mechanisms
frequently consumes significant energy to detect the target again. In our study,
in order to reduce the miss ratio of target, we propose to form lookahead spanning trees in the predicted trajectory. Additionally, as a recovery mechanism we
propose to increase the number of waked up nodes gradually. As the probability
of missing the target decreases, the need for recovery mechanisms also decreases
leading to reduction in energy consumption.
To summarize, an intended contribution of our work is that we propose multi
cluster version of the dynamic spanning tree based approach for very fast moving targets. The other intended contribution is that we compare the proposed
tracking algorithm with generic cluster and generic spanning tree based tracking
approaches. In addition, we also compare the proposed algorithm with the multi
cluster based tracking algorithm [8] for target miss ratios and energy consumption. Lastly, by decreasing the usage of recovery mechanisms we also decrease
the energy consumption leading to increment in network lifetime. The rest of
this paper is organized as follows: In Section 2, the proposed tracking algorithm
is explained in detail. The performance evaluations obtained from simulation
results are presented in Section 3 where comparisons with few other algorithms
are also done. Finally, conclusions and future works are given in Section 4.
2
The Algorithm
The Dynamic Multi Spanning Tree Based Tracking Algorithm(DMSTA) aims to
track very fast moving targets by decreasing the ratio of missing the target. In
order to decrease the ratio of target misses, the algorithm forms lookahead trees
along the predicted target direction. The algorithm is designed in a distributed
manner. The pseudocode of the proposed algorithm is given in Algorithm1.
Lecture Notes in Computer Science: Authors’ Instructions
3
Algorithm 1 The pseudocode of the DM ST A for nodei
1: if I am an ordinary node
2:
if the target is detected
3:
select a rootj node
4:
if i am root msg is received
5:
become member node
6:
if hop count value is greater than zero
7:
send i am root msg to neighbor nodes
8: if I am a member node
9:
if the target is detected
10:
send information msg to leaderj node
11:
if you are root msg is received
12:
if hop count is greater than zero
13:
send you are root msg to the neighbor node towards
14:
the direction of the target
15:
else if hop count is equal to zero
16:
become root node
17:
if tree hop count is greater than zero
18:
send you are root msg to the neighbor node towards
19:
the direction of the target
20:
if raymond root msg is received
21:
become root node
22: if I am a root node
23:
if information msg is received
24:
calculate location, speed and direction of the target
25:
calculate the number of spanning tree to be formed
26:
(tree hop count)
27:
send you are root msg to the node towards the direction of the
28:
target
The algorithm contains root election, initial tree formation, dynamic root
management, lookahead tree formation and recovery mechanism phases. In this
section, these phases are explained in detail.
2.1
Root Election and The Initial Tree Formation
When the target enters the tracking area, nodes in active state and closer to the
target detect it and form the initial tree by first selecting a node to be the root
4
1
Aysegul Alaybeyoglu,
1
Aylin Kantarci,
2
Kayhan Erciyes
of the tree. We use a two phase timer based election algorithm which selects
the node closest to the target as the root node. In this algorithm, each node i
that detects the target, sets a timer which varies in accordance with the node’s
received signal strength. The higher the received signal strength, the smaller
the timer value is set. The node whose timer expires first is the closest node
to the target. Each node waits for expiration of its timer and does not send
any messages to its neighbors until the timer expires. If a node does not receive
any candidate messages until the timer expires, it becomes a root candidate and
sends candidate(rss, id) message to its neighbors including its rss and its own
id. Otherwise, it gives up and selects the node that sends the candidate message
as the root node. Since the nodes may not be in one hop communication range of
each other, two or more root candidates may exist after the first phase. For that
reason, after the first phase, each root candidate node i, sets a second timer. If
until the second timer expires, the root candidate node i, receives a candidate
message with higher rss value, it gives up the candidacy. As a result, all root
candidate nodes give up, but one with the highest rss value becomes the root
node.
2.2
Dynamic Root Management
After the initial tree is formed, the root node collects all sensed data from the
nodes in its tree. As the target moves, the closest node to the target in the tree
also changes. From the received signal strength values, the current root node notices that there is a node closer to the target than itself and sends raymond root
message to the closest node. As a result, in the proposed algorithm, the root
node changes dynamically so that the node closest to the target always becomes
the root node and target direction is predicted more accurately. In sequence, the
tree self organizes itself by rearranging the affected links. We inspired from Raymond’s algorithm, a token based algorithm for mutual exclusion on distributed
system to reconfigure the tree [9]. It can be seen from the Fig. 1 that when the
root node changes, the path from member nodes to the root node also dynamically changes.
Fig. 1. Root Changes Dynamically.
Lecture Notes in Computer Science: Authors’ Instructions
2.3
5
Lookahead Tree Formation
Upon detection of the targets, the root node calculates the location, speed and
the direction of the target. In accordance with the calculated speed, the root
node defines the number of spanning trees to be formed and starts lookahead
spanning tree formation process. Fig. 2 illustrates an example operation of the
lookahead tree formation process.
n12
n16
n17
n13
n18
n10
n14
n7
n4
n9
n2
hop_count -hop_count =0
n5
n1
n3
hop_count -hop_count !=0
n11
hop_count -hop_count !=0
n15
n6
n8
you_are_root (hop_count , trajectory)
i_am_root (idi)
message
message
Fig. 2. Lookahead Spanning Tree Formation
As it can be seen from Fig. 2, current root node n1, collects sensed data via its
spanning tree and calculates the location, speed and the direction of the target.
n1 sends you are root message including hop count and trajectory parameters to
the neighbor node n5 that is closest node to the target’s direction. The hop count
parameter defines the number of spanning trees to be formed along the direction
of the target. n5 forms its spanning tree and decreases the hop count value. If
hop count value is still greater than zero, it continues to send the you are root
message received from node n1, to the neighbor node n11 that is closest to the
target’s direction. This process continues until the hop count value becomes zero.
2.4
Recovery Mechanism
In case the target moves at high speed, it can pass by a group of nodes very
fast without being detected. Therefore as the target increases its speed, the
probability of missing that target also increases. A recovery mechanism is needed
when the target missing occurs. A basic solution to recapture the missing target
is to wake up all sensor nodes in tracking area which results in very high energy
consumption.
6
1
Aysegul Alaybeyoglu,
1
Aylin Kantarci,
2
Kayhan Erciyes
In the proposed algorithm, when the current root node notices the target
missing, it enlarges its tree by flooding alarm message to its n hop neighbors.
When a node receives this message, it sends a reply message including received
signal strength value if it has detected the target. Current root node continues to
enlarge its tree until it determines the target’s location. The recovery mechanism
used in the proposed algorithm is given in Fig. 3.
n+1
n
Fig. 3. Recovery Mechanism
Tracking very fast moving target with the existing approaches results in
high miss ratios. Due to the high miss ratios, existing tracking approaches use
recovery mechanisms much more frequently than the proposed algorithm which
also results in more energy consumption.
3
Simulation Results
We implemented our tracking algorithm in ns2 simulator [?] version 2.31. Our
mobility pattern of the target is random waypoint model which is supported by
ns2. The speed of the target is varied to measure the detection capability of our
algorithm under different mobility conditions. The speeds are chosen from 30
m/s to 100 m/s.
In order to evaluate the performance of DMSTA, four different algorithms
are implemented: The Generic Dynamic Cluster Based Tracking Algorithm (GDCTA) which constructs a cluster dynamically upon detection of the target, the
Generic Dynamic Spanning Tree Based Tracking Algorithm (GDSTA) which
constructs the trees dynamically upon detection of the target, the Dynamic
Multi Cluster Based Tracking Algorithm (DMCTA)which pre-constructs clusters along the trajectory of the target and finally our Dynamic Spanning Tree
Based Tracking Algorithm (DMSTA).
Lecture Notes in Computer Science: Authors’ Instructions
7
We plotted the target’s movements against speed to illustrate the tracking
accuracy of our algorithm. As shown in Fig. 4.a and Fig. 4.b, calculated coordinates are very approximate to the real coordinates when the speed is varied
between 30 m/s to 40 m/s and 50 m/s to 60 m/s.
(a) Mobile Scenario.
(b) High Mobile Scenario.
(c) Very High Mobile Scenario.
(d) Extremely High Mobile Scenario.
Fig. 4. Actual and Calculated Traces for Dynamic Multi Spanning Tree Based
Target Tracking Algorithm
When the speed is varied between 70 m/s to 80 m/s and 90 m/s to 100
m/s, the difference between real positions and calculated positions is greater
but it is very important that the trajectory is preserved under very high mobile
conditions as shown in Fig. 4.c and Fig. 4.d.
We measured the miss ratios of GDCTA, GDSTA, DMCTA and DMSTA
against very high speeds in Fig. 5. We obtained from the simulations that the
8
1
Aysegul Alaybeyoglu,
1
Aylin Kantarci,
2
Kayhan Erciyes
Fig. 5. Comparison of Miss Ratios.
proposed algorithm gives better results than the generic approaches against target misses.
When we compared the proposed algorithm with DMCTA, the best performance is obtained with DMSTA in terms of target misses for fast moving targets.
The reason for this is the higher probability of catching the target with one of
the formed lookahead spanning trees.
In Fig. 6, we compared the algorithms in terms of energy consumption. As the
target misses increase, in order to recapture the missing target, the need and
usage of recovery mechanisms also increase leading high energy consumption.
As it can be seen from the figure, the highest energy consumption is obtained
in GDCTA algorithm because of the frequently usage of recovery mechanisms.
GDSTA and DMCTA follows GDCTA in terms of miss ratios. Inspide of lookahead tree formation, the proposed algorithm DMSTA results in the least energy
consumption due to the least usage of recovery mechanism.
Lecture Notes in Computer Science: Authors’ Instructions
9
Fig. 6. Comparison of Energy Consumptions
4
Conclusion
In this paper, a new distributed tracking algorithm namely DMSTA is proposed
for very fast moving targets at speeds up to 100 m/s which is about three times
higher than the highest speed of objects that can be tracked by sensor networks
reported in literature. The aim of our algorithm is to decrease the miss ratios
as well as energy consumption. The proposed algorithm is compared with three
different tracking algorithms in terms of miss ratios and energy consumption and
it has been shown by the simulation results that, for a range of speed values from
30 m/s to 100 m/s, DMSTA gives the best performance in terms of both miss
ratio and energy consumption. DMSTA performs well in miss ratios because of
the lookahead tree formation. As the miss ratio decreases, the usage of recovery
mechanisms also decreases resulting in less energy consumption. We also showed
by simulations that the proposed algorithm performs well by constructing paths
very close to the original motion of the target at high speeds.
References
1. B. Thorstensen, T. Syversen, T. Walseth and T. Bjornvold, ”Electronic Shepherd:
A Low-Cost, Low-Bandwidth, Wireless Network System”, Proc. Second Intl. Conference on Mobile Conference on Mobile Systems, Applications and Services, pp.
245-255, 2004.
2. P. Zhang, C.M. Sadler, S.A. Lyon and M. Martonosi, ”Hardware Design Experiences
in ZebraNet”, SenSys04, ACM, pp. 227-238, 2004.
3. W. Yang, Z. Fu, J. Kim, and M.-S. Park,”An Adaptive Dynamic Cluster-Based
Protocol for Target Tracking in Wireless Sensor Networks”, in Proc. of. WAIM 07,
2007, pp. 157-167.
10
1
Aysegul Alaybeyoglu,
1
Aylin Kantarci,
2
Kayhan Erciyes
4. W. Chen, and J. Hou, ”Dynamic Clustering for Acoustic Target Tracking in Wireless
Sensor Networks”, IEEE Transactions on Mobile Computing, vol. 20, 2004, pp. 258271.
5. R. R. Brooks, P. Ramanathan, and A. M. Sayeed, ”Distributed Target Classification
and Tracking in Sensor Networks”, IEEE Signal Processing Magazine , vol. 91, 2002,
pp. 1163-1171.
6. S. Suganya, ”A Cluster-Based Approach for Collaborative Target Tracking in Wireless Sensor Networks” , in Proc. of ICETET 08, 2008, pp. 276-281.
7. W. Zhang, and G. Cao, ”DCTC: Dynamic Convoy Tree-Based Collaboration for
Target Tracking in Sensor Networks”, IEEE Transactions on Wireless Communications, pp. 1689-1701, 2004.
8. Alaybeyoglu, A., Erciyes, K., Kantarci, A., Dagdeviren, O., ”Tracking Fast Moving
Targets in Wireless Sensor Networks”,IETE Technical Review, 2010
9. R. Chow, T. Johnson, ”Distributed Operating Systems and Algorithms”, Addison
Wesley, 1997.
© Copyright 2026 Paperzz