![](bg2.png)
in terms of power consumption and communication
complexity. Consequently, these approaches are not
efficient because they require extensive inter-
communication among neighbor nodes. In
comparison with the distributed version of Kalman
Filter in literature (Rao, et al., 1991; Olfati-Saber,
2007; Alriksson, et al., 2007; Olfati-Saber, et al.,
2005; Cattivelli, et al., 2008; Hashemipour, et al.,
1998), our version of the distributed Kalman Filter
simplifies computational burden and reduces inter-
node communication. Thus, the total power
consumption in the entire sensor network is lower
than that reported elsewhere in the literature.
Our approach is different from the above work in
the sense that the Kalman Filter is implemented in a
distributed fashion across the WSNs. At a given
instant, only one master node runs the Kalman Filter
using the measurement inputs from its neighbors and
shares the estimated knowledge with the subsequent
master node. The neighbors within a certain distance
from the target measure the distance to the target,
and transmit measurements to the master node. On
one hand, the procedure significantly reduces the
communication costs among the neighbor nodes in
comparison with the algorithms proposed in (Rao, et
al., 1991; Olfati-Saber, 2007; Alriksson, et al., 2007;
Olfati-Saber, et al., 2005; Cattivelli, et al., 2008;
Hashemipour, et al., 1998). On the other hand, since
the master node alone executes the Kalman Filter
and the neighbor nodes only perform measurement
functions, the complexity of the WSN is greatly
reduced.
Another contribution of this paper is that the
master node determines the direction and velocity of
the intruder and wakes up appropriate sensor nodes
in the direction of the target travel. As the target
moves into the sensing range of a sensor node, it is
already activated and is ready to take measurements.
Whereas the other nodes that are far away from the
target are automatically turned off to save energy.
The master node also decides to wake up sufficient
nodes to take measurements. By knowing the
maximum target’s velocity, the boundary nodes of
the sensor field are activated in round robin fashion
discussed in (Watfa, et al., 2006b) to save energy.
Unlike other approaches mentioned above, we
do not make an assumption about the linear
movement of the target. In this paper, the distributed
Kalman Filter is proposed to estimate the position of
the target. This approach is validated through
simulation examples and the results are compared
with those represented in literature. We show the
main contribution, the approach, validations, and
comparison between our method and the previous
work on distributed Kalman filtering. The algorithm
was also able to track the target with random
directions with acceptable estimated results. The
estimation results showed that the model is robust to
measurement noise and the change in velocity. The
estimated knowledge of the Kalman Filter including
system state and covariance matrix is passed directly
to the subsequent master node where the Kalman
Filter is run. Consequently, the performance of the
distributed Kalman Filter is as good as that of the
centralized Kalman Filter.
The rest of the paper is organized as follows:
Section 2 discusses the algorithm in details. In
section 3, we show the numerical simulation.
Section 4 and 5 are discussion and conclusion.
2 ALGORITHM
2.1 Problems and Assumptions
A sensor field is densely deployed with sensor
nodes. It is assumed that each node has
omnidirectional sensing capability to measure the
distance between the target and itself. Moreover,
every node knows its coordinates in the sensor field,
and all nodes are stationary. Initially, all the nodes
except those at the boundary of the monitored area
are assumed to be in sleep mode. Assuming that
there is an intruder entering the sensor field with an
unknown nonlinear trajectory and a known
maximum velocity, the problem is to track the
position of the intruder accurately. When a target
moves in the sensor field, the nodes close to the
target will automatically activate and sense the
target.
All sensing nodes are within one communication
hop from the master node. The trilateration
algorithm requires that every point in the field is
covered by at least three sensor nodes.
A node can be either the master node or a
measurement node. Nodes take measurements and
sends data to the master node if they are actively in
the sensing region. Concurrently, the master node
collects data from its neighbors, running estimation
algorithms and broadcasting the information of the
target to its neighbors, including the target’s current
coordinates and direction. Depending on the
information from the master node, the neighbor
nodes around the target automatically turn off when
they are not in the region of activation R around
which is defined as the following .
The target, represented by symbol shown in
Figure 1, is moving in horizontal direction. The
DISTRIBUTED KALMAN FILTER-BASED TARGET TRACKING IN WIRELESS SENSOR NETWORKS
55