- School of Science and Computer Engineering

advertisement
Master Thesis Proposal
Applying image processing techniques to
simulate a self organized sensor network
for tracking objects
By
Sam Tran P.
School of Science and Computer Engineering
University of Houston – Clear Lake
12/2004
Committee members and signatures
Approved by:
Date:
------------------------------------------------------------Advisor: Dr. T. A. Yang
------------------------------------------------------------Committee member: Dr. L. Shih
------------------------------------------------------------Committee member: Dr. G. C. Collins
Deans
-------------------------------------------------------------------------
1
Abstract
This thesis aims to simulate a distributed sensor network for tracking
objects, and to evaluate the performance and security of tracking algorithms
in the implemented simulation. A distributed sensor network is composed of
computer nodes that communicate wirelessly between each other, and with
the base station. Based on the node positions, the base generates a
geographic image about the network, and uses the network image to
organize the network for the purpose of tracking objects entering the
covered area of the sensor network. Existing self-organization algorithms for
sensor networks tend to be inaccurate, and, in order to improve accuracy,
may consume lots of energy, which is typically not affordable in wireless
sensor nodes. I propose the development of new tracking methods, by
applying techniques of image processing, such as border detection and
skeletonization, to improve the accuracy of tracking objects in a sensor
network. The proposed method, and some of the existing methods, will be
evaluated in the simulated sensor network. Both performance and security
issues will be investigated.
2
1. Introduction
A wireless sensor network is a network of wireless sensor nodes. Each
node is a computer with attached sensors that can process, exchange
sensed data, as well as communicate wirelessly among themselves to
perform various tasks.
Since the network topology frequently changes due to environment
conditions and multi-hop communication is required, reconfigurable self
organization (including routing and sensor deployment) is the most
important aspect of designing a wireless sensor network.
Besides, each sensor node is a low performance computer with a tiny
battery. Hence, minimizing communications among the nodes is one of the
essential requirements in designing a network of wireless sensor nodes.
2. Related work
The current self-organization algorithms are based on the LEACH (Low
Energy Adaptive Clustering Hierarchy) algorithm, in which the nodes are
grouped into a hierarchical tree [1] (as illustrated in Figure 1). The node at
the root level represents the base station, the leaf level consists of sensor
nodes, and the middle is replay nodes which collect, examine, compress, or
encrypt the sensed data and then hierarchically forward the processed data
to the base.
3
Figure 1: Nodes are grouped into a hierarchical tree in LEACH
The LEACH Algorithm
First, a random node broadcasts to its neighbors its intent to be the
head of a cluster. Each neighbor might receive many invitations to join the
clusters. However, it selects which cluster it wants to be a member of. Once
all nodes are organized, each head makes a schedule for each of its member
nodes. And then, these heads also belong to higher level clusters and so on
until they hit the base at the top level (the root). [2]
Tracking
4
As shown in Figure 1, in the hierarchical tree, the lower the level the
higher the number of nodes that level has. Normally, the base at the top
level does not activate all the nodes in the tree, but just nodes down to a
certain lower level. When an active node senses an object, it will activate all
nodes in its cluster, collect the sensed data, examine (or compress) it, and
forward it to the base.
Weaknesses of LEACH
Inaccuracy: Normally, only the heads are activated. That is, only a few
points in the sensing area are active at a time. It leads to inaccuracy in
tracking objects. To increase accuracy, the network must activate nodes at
low levels. However, lower levels contain more nodes. The network will
consume more energy when more nodes are activated.
Another issue related to inaccuracy is that multiple sensors may sense
the same data simultaneously. Thus, they increase the overhead of the head
being required to process redundant data. (To deal with this challenging
issue, some algorithms have been applied to detect the overlapped sensing
among nodes in a cluster [3].)
Waste of Energy: When detecting an object, all nodes in a cluster are
activated. Also, when the head dies, all nodes in the clusters have to re-elect
another head. These operations involve a lot of communication among the
nodes, and thus incur extra demand for battery power.
5
3. Proposed method
The proposed algorithm uses the border detection and skeletonization
method to handle the task of tracking objects in the wireless sensor
network.
Step 1: Send position information to the base:
When a large number of sensor nodes are deployed, each of them
needs to send its own position back to the base station. Figure 2 shows how
the base and the sensor nodes may interact to have this task accomplished.
Base
Ask
Return
Figure 2: Nodes sending location information to the base
Step 1a: The base sends its ID to its neighbors and asks them about
their IDs and locations.
Step 1b: After receiving the information, the base will assign its ID as
the father for the nodes that have responded, and require them to ask
6
their neighbors about IDs and location information and then forward
the information back to the base.
Step 1c: At each of the neighbors, step 1a repeats. The recursion
continues until all nodes are reached. At that time, the base gets
information (positions and IDs) of all nodes.
Step 2: Assign tasks to nodes (Figure 3):
Base
Sub-skin
Skin
Skeleton
Figure 3: Image processing to assign task to nodes
Step 2a: After getting all information about the network, the base
builds a geographic image of the network. The skeleton of the image
forms the main route. The border acts as the skin (all nodes on the
skin are in active state). If the distance from the skin to the skeleton is
7
too far (out of the coverage zone of a node), sub-skin layers are
created to replay messages from the skin to the skeleton (the main
route).
Step 2b: The base then assigns different functions to various nodes in
the sensor network. The nodes are assigned respectively as skin
(active), sub-skin (replay), skeleton (main route), or reservation
(sleep). Each node has a father, possibly one or more sons, and some
neighbors, which include the neighboring nodes that are not in sleep
mode and not the father or sons.
Step 3: Track objects:
We assume that tracked objects come from outside of the sensing
zone. Normally, only skin nodes are active. When a node detects an object,
it sends the signal to its father and wakes up the appropriate nodes (father
or sons). The father will transfer the message to the main route (the
skeleton) to get to the base. At the appropriate node, after switching to
active, it wakes up the neighbors to be active to prepare for sensing the
object (because the object could move toward the neighbors).
The process continues recursively, and the sensor network continually
monitors the object. Information about the object is continually forwarded to
the base, where the information is updated. The sub-skin nodes will
8
automatically switch to replay status after a specific period of idle time,
when nothing is sensed.
Step 4: Maintenance:
To keep information about the sensor network up-to-date, each node
periodically checks the availability of its neighbors by sending a testing
message to them. If a node finds out that one of its neighbors is dead, it
reports the dead neighbor’s ID to the base. The base updates the
configuration, and re-examines the network map to decide which method
replacement or re-assignment should be applied to keep the network
operational (as discussed in step 2 above).
It is noted that security issues will need to be resolved in maintaining
the effective operations of the sensor network. Take the periodic
maintenance (step 4) as an example. How can the base prevent a malicious
node from sending false information with the purpose of disabling the
network?
4. Thesis plan
The thesis will start from spring 2005 and end in summer 2005.
Step 1 (2 weeks)
Install and configure the simulation tool (TOSSIM or NS-2).
Step 2 (2 weeks)
9
Generate the geographic image from simulation node locations and its
coverage zone.
Step 3 (4 weeks)
Build the skeleton, the skin (border), and the sub-skins from the
image.
Step 4 (7 weeks)
Use the tool to simulate the algorithm, and investigate its performance
and security impact on the sensor network.
Step 5 (5 weeks)
Test the method. Simulate LEACH algorithm and then compare the
accuracy and power consumption of the algorithm to the LEACH. The
expected outcome is that the proposed algorithm consumes less energy and
gives more exact results.
10
References
1. Manish Kochhal, Loren Schwiebert, and Sandeep Gupta (2004).
“Integrating Sensing Perspectives for Better Self Organization of Ad
Hoc Wireless Sensor Networks”. JOURNAL OF INFORMATION SCIENCE
AND ENGINEERING 20, 449-475 (2004). Retrieved 11/7/2004 from
http://newslab.cs.wayne.edu/jise.pdf
2. Wendi Rabiner, Anantha Chandrakasan, and Hari Balakrishnan (10/2004).
“Energy-Efficient communication protocol for Wireless Microsensor
Networks”. THE 33RD HAWAII INTERNATIONAL CONFERENCE ON
SYSTEM SCIENCES (Oct 20, 2004). Retrieved 11/12/2004 from
http://faculty.cs.tamu.edu/dzsong/teaching/fall2004/netbot/John_G.p
pt
3. Himanshu Gupta, Samir R. Das, Quinyi Gu (2003). “Connected Sensor
Cover: Self-Organization of Sensor Networks for Efficient Query
Execution”. THE FOURTH ACM INTERNATIONAL SYMPOSIUM ON
MOBILE AD HOC NETWORKING AND COMPUTING (MOBIHOC) (2003).
11
Retrieved 11/10/2004 from
www.cs.sunysb.edu/~hgupta/ps/coverage.ps
12
Download