Transcript 下載/瀏覽Download

Tree-Based Object Tracking Without
Mobility Statistics in
Wireless Sensor Networks
指導教授：李育強
報告者 ：楊智雁
2010/12/24
南台科技大學
資訊工程系
Outline
1
Introduction
2
Query Cost Reduction
3
2
Analytic Mobility Profiling
4
Experimental Results
5
Conclusion
1. Introduction
 Rapid progress in wireless communications and
micro-sensing MEMS technology have enabled the
deployment of wireless sensor networks
 Object tracking is an application of WSNs where
the presence of particular mobile objects can be
detected by nearby sensors
 A challenge is to coordinate these sensors to make
the tracking process more accurate, dependable
3
1. Introduction (c.)
 All nodes are organized into a tree structure rooted
at the sink
 When a sensor detects a target object entering into
its duty area, it sends an Enter message toward the
sink to create or update the associated query path
 Such a tree-based object tracking approach incurs
two types of message costs
4
1. Introduction (c.)
 update cost and query cost
5
2. Query Cost Reduction
 The update cost of T is given by
u(T ) 
 ((u, v)  d
( u ,v )E
T
(u, v ))
 The query cost is
q(T ) 

uL ( T )
6
qu  d T ( S , pT (u )) 
q
uL ( T )
u
 dT ( S , u)
2. Query Cost Reduction (c.)
 Average node level (ANL)

ANL(T ) 
uV
d T ( u, S )
V
 Average reporting length (ARL)

ARL (T ) 
7
( u ,v )E
d T ( u, v )
E
2. Query Cost Reduction (c.)
 The first considers turning an intermediate node
into a leaf by rewiring all its children to its parent
 The second technique attempts pulling leaf nodes
up one level
8
3. Analytic Mobility Profiling
p 
0
i

Ai
N
j 1
Aj
 Obtain state transition probabilities for X(t)
X (t )  i
 Let M(m; n) be the transition probability matrix
for times m and n, where m < n
9
3. Analytic Mobility Profiling (c.)
p ( n )  p ( m )  M ( m, n )
10
4. Experimental Results
 Two mobility models were used to drive object's
movements, Random Waypoint and GaussMarkov
 100 sensor deployments were randomly generated
11
4. Experimental Results (c.)
12
4. Experimental Results (c.)
 Node Level and Reporting Length
13
4. Experimental Results (c.)
 Update Cost
14
4. Experimental Results (c.)
15
4. Experimental Results (c.)
 Query and Overall Costs
16
4. Experimental Results (c.)
17
5. Conclusion
 Two heuristic designs for the OMRT problem,
DAT and MST, have been discussed
 The results show that in case of MST, the proposed
analytic profiling outperforms statistical profiling
 In case of DAT, the proposed analytic profiling
performs the same as the statistical profiling
18
南台科技大學
資訊工程系