### Sanaz Khakpour: D.istributed clustering algorithm-ppt

```A Distributed Clustering Algorithm for
Networks
Dr. Khalil El-Khatib, Dr. Richard Pazzi, Sanaz Khakpour
- Introduction
- Related Work
- Algorithm Features
- Algorithm Overview
- Algorithm Description
- Conclusion
Introduction
• VANETs are network of autonomous mobile nodes that
communicate with each other without any fixed
infrastructure.
• VANETs are large-scale networks and dividing the network
into smaller clusters in such dynamic environment is an
Related Work
• MANET and WSN clustering algorithms do not work
properly in VANET environment.
• The most important challenge in clustering algorithms
that most of the protocols are trying to solve are:
 Optimal cluster management in VANET’s highly
dynamic environment.
 Increasing cluster stability (MDMAC, SBCA)
 Prevention of frequent cluster changes
 Increasing cluster head availability (SBCA)
 Increasing cluster lifetime by using appropriate
mobility metrics (DCA, MDCAM, DMAC, SBCA,
MOBIC, …)
Special features of the proposed algorithm
• A cluster-based target tracking algorithm
•
robust and stable clusters
•
low delay and overhead for electing new cluster
•
-Introduction
- Related Work
- Algorithm Features
- Algorithm Overview
- Algorithm Description
- Conclusion
Assumptions and Definitions
• The proposed clustering algorithm was designed for
vehicle tracking in VANETs.
• This algorithm assumes that vehicles have front and
rear cameras and can detect visual features of a
target.
• A central entity such as a police station is seeking
help to find a specific target. This entity is called
Control Centre (CC) and is a node located in multihop communication distance from target.
Tracking Failure Probability Metric (TFP)
Assumptions:
• All vehicles are aware of their location and velocity by
using a GPS device.
• The location of a target is unknown; But can be calculated
by visual processing.
• To calculate TFP between a vehicle A and the target
vehicle T at time t, we need to have the distance between
node A and T and their speed vector at that time.
• We define a value called Valid Distance Range (VDR),
which is used to normalize the distance between any node
and target.
Tracking Failure Probability Metric (TFP)
• The normalized distance is calculated as follow:
(1)
=

• By using velocity vectors of vehicles we can
differentiate between nodes moving in the same
direction and nodes moving in opposite direction.
is defined as:
(2)
•
=
We use a value called Valid Velocity Range (VVR) in
order to normalize the value of velocity vectors.
Tracking Failure Probability Metric (TFP)
• V and V Are normalized velocity vectors of vehicle A and
target T respectively.
(3)
=

(4)  =

• Two values α and β are defined as Distance and speed
Efficiency Factors. These values are coefficients of distance
and velocity to control efficiency of these metrics of each
vehicle.
• The TFP of node A at time t is calculated as in the following
formula. The lower TFP indicates higher priority to become
(5)
() = 100 * ( + β  −  )
-Introduction
- Related Work
- Algorithm Features
- Algorithm Overview
- Algorithm Description
- Conclusion
-Introduction
- Related Work
- Algorithm Features
- Algorithm Overview
- Algorithm Description
- Control Centre Functions
- Initialization Phase
- Cluster Management Phase:
o Cluster Members Functions
- Tracking Phase
- Conclusion
Control Centre (CC)
• CC broadcasts a “Target Tracking Request Message”
(TTRM) to the entire network with target vehicle’s visual
information.
• When CC receives “Response Message” from any
vehicle that has located the target, it will stop
• At any point later, if the CC stops receiving any
information from the cluster head regarding the
specified target (after a pre-defined time interval) it will
assume the cluster no longer exists and it will start
broadcasting target’s information again in the network.
Initialization Phase
• Any vehicle that receives a TTRM message from the
Control Center (CC) and which can detect the target
responds to CC and starts the initialization process.
• OBNs start broadcasting “Request Messages” to their
neighbors and receive “Response Messages” from
them. OBNs also check the TDV field of the response
messages.
• OBNs calculate their TFP. This value displays which
vehicle has closer movement pattern to target and is
more appropriate to be the cluster head.
Initialization Phase
• Cluster members are divided into 2 groups: level-1 (OBNs)
and Level-2 (NNs).
• Member nodes are connected to cluster instead of
• Initialization phase might be repeated only if there is not
any cluster members available and the cluster is
destroyed.
• The purpose of our design is to avoid switching to
Initialization Phase from Cluster Maintenance phase.
• After this phase the initial cluster is created and the
Cluster Maintenance Phase
• CH is responsible of managing the cluster by sending
request messages at every time intervals to find new cluster
members.
• the cluster head calculates its own TFP every, and
compares it with other values in the neighbour list to check
if it is still a valid CH. If not it will send a “Resign Message”.
• A “Safe Threshold” is defined because the TFPs are
changing so quickly and we do not want to change CH so
frequently.
• vehicles moving in opposite direction of the target are not
supposed to join cluster because these nodes are unstable
cluster members and will decrease cluster’s stability.
Cluster Maintenance Phase
(Cluster Members Functions)
• OBNs calculate their TFP every  time interval and send it in
RPM to their neighbors. Also, OBNs store the TFP of other
nodes in their neighbor list.
• If member nodes receive a RSM they are responsible to find
a node with lowest TFP value in their neighbor list and select
it as CH.
• Also, if a member node does not receive any kind of
message after a defined time interval, it assumes to be out
of cluster borders and will try to send its information directly
to CC (if possible).
Tracking Phase
• Tracking includes taking continuous video of target and
sending video data and location information of target to
CC during specified time intervals.
• CMs send their data to CH and CH is responsible to inform
• In case CM goes out of cluster boundaries (and does not
CC.
-Introduction
- Related Work
- Algorithm Features
- Algorithm Overview
- Algorithm Description
- Conclusion
Conclusion
• Introduced
algorithm
aims
to
improve
cluster
performance by making stable and long living cluster.
• The stability of this algorithm is mostly because of adding
candidate cluster members which are highly probable of
detecting target in near future.
• The TFP value is used as an evaluation value to compare
movement pattern of vehicles with target.
• The
idea
of
distributed
introduced by use of TFP.
cluster