2012 8th IEEE International Conference on Distributed Computing in Sensor Systems

The Design and Implement of Acoustic Array Sensor

Network Platform for Online Multi-target Tracking
Yuanshi Li, Zhi Wang*, Shuguo Zhuo and Jie Shen Shengsheng Cai, Ming Bao and Dahang Feng
State Key Laboratory of Industrial Control Technology Institute of Acoustics
Zhejiang University Chinese Academy of Science
Hangzhou, Zhejiang Province, China Beijing, China
wangzhizju@gmail.com Baoming@mail.ioa.ac.cn

Abstract—We present the design, implement and evaluation of a information from fusion center to sensor nodes as necessary
sensing platform for on-line multi-target tracking based on instructions.
acoustic array networks, named Integrated Acoustic Array
Sensor Network Tracker (IAASNT). To provide on-line multi-
target tracking service, a well-designed system structure is
proposed, composed by supporting components and associations
between each part. Among these, IAASNT’s multi-level low-
power management and integrated tracking frame set it different
from other related platforms. The integrated tracking frame is
the core of the system and has been carefully designed, to achieve
a self-acting tracking service. Finally, a series of experiments on
system have been done to evaluate the performance of IAASNT.
The tracking experiments on system show a perfect tracking
performance in both noise-free and noisy environment, and the Figure 1. Typical acoustic array sensor network
tracking precision can be within 5.8m in 300*300m area.
In this paper, we present the design, implement and
Keywords-acoustic array networks; system design; integrated evaluation of Integrated Acoustic Array Sensor Network
target tracking; low-power management; tracking experiments Tracker (IAASNT), a platform for on-line multi-target tracking.
Each sensor node in IAASNT is developed by embedded
I. INTRODUCTION system and can get the bearing estimation by sampling and
Target tracking attracts much attention for its great processing acoustic signal from targets. Comparing with the
application in military, scientific and civil. Since the past classical tracking system such as radar, acoustic array sensor
decades, the related theory has achieved great development and network systems have advantages including passive measuring
many target tracking systems have also been developed. In is low-power consuming and well stealthy, and network frame
applications, especially in military application, however, active can achieve more general and robust service. IAASNT’s
sensors like radar are easily detected, located and destroyed by advantages also lies on its well-designed system structure and
enemies. So this kind of system does not fit for this special the implement of supporting components in the structure.
application. As passive sensor technique and wireless sensor Among these, components for integrated target tracking and
networks (WSN) develop, passive sensor networks provide a multi-level low-power management are the highlights of the
new era for target tracking and acoustic array sensor networks system distinguish it from other related systems. The challenge
is a nice try among these. However, for lacking of study in a in designing the tracking component is how to combine
system view on this fields and the limitation of processing classical tracking model and the supplement for the special
capacity on sensors, little work has provided an integrated on- characteristics in this system and the integrated tracking frame
line tracking service under acoustic sensing platform. We make we proposed gives a feasible solution. Based on this frame, we
efforts to achieve a sensing platform for on-line multi-target do efforts on modules such as initial state estimation
tracking based on acoustic array sensor networks. combining with track initiation, node selection. To evaluate the
performance of system, we perform experiments on tracking.
Figure 1 shows a typical acoustic array sensor networks for
target tracking application. Several sensor array nodes locate in Our IAASNT system is a nice try for the target tracking
the area. When a target occurs, the sensor nodes can get application in WSN including innovations both in theory and
measurements about target motion such as bearing measuring practice. The contributions mainly lie on the following points:
by utilizing the phase difference between array elements. Then • Firstly, the system structure of IAASNT is well-designed
measurements will be uploaded to the data fusion center and and a series of approaches are proposed for components in
target tracking estimation can be obtained here. Sometimes, a this structure. All these work can be easily referenced by
feedback structure is needed to transfer some important scientists and engineers to construct similar systems.

This work has been partially supported by joint ANR-NSFC Quasimodo project
(ANR 2010 INTB 020601 and NSFC 61061130563), SKLICT project (ICT1103)
and KLWSNC CAS project (WSNC2011001). *Corresponding author

978-0-7695-4707-7/12 $26.00 © 2012 IEEE 323

DOI 10.1109/DCOSS.2012.45
• Secondly, IAASNT has realized on-line multi-target III. PLATFORM OVERVIEW
tracking with relatively high precision, which is a great The IAASNT system provides an integrated on-line multi-
challenge to tracking systems in acoustic sensing systems, target tracking service. To achieve the goals, we design a novel
especially in acoustic array sensor networks. system structure as shown in Figure 2. The structure consists of
• Finally, many experiments have been arranged to test the 3 layers including application layer with target tracking
performance of the system and from which practical component, middleware layer with communication,
experience and theory are well combined. synchronization, signal process and node localization
This paper is organized as following. Section 2 discusses components, hardware layer with hardware of sensor node for
the related work on acoustic sensing system. Section 3 and a cross-layer with low-power component.
describes the system structure of Integrated Acoustic Array
Sensor Network Tracker and components in this structure. On the basis of the hardware of sensor node, signal process
Section 4 introduces the tracking experiments and analysis. and communication components achieve basic function of
Finally, Section 5 summarizes the system and gives the future sensor network, that is sensing and data transmission. Then
work. node localization and time synchronization components realize
a practical sensing system for providing determinate position
and time. All the above supporting components make tracking
II. RELATED WORK
service to be a simple multi-input multi output (MIMO) model.
In this section, we give an overview on related acoustic In addition, a multi-level low-power management is also one of
target sensing platforms. The design of acoustic target sensing the most meaningful works in this system.
platforms attracts much attention from academe, industry and
military. A number of projects have been accomplished by
research institutes. Acoustic ENSBox [3] and VoxNet [4]
designed by Grid etc. are distributed acoustic sensing platforms
based on ARM. The work is mainly on self-localization by
using of acoustic range and bearing estimation. Another
famous work done by UCLA is acoustic sensor networks for
woodpecker localization [5]. By using several acoustic arrays
which consists of four microphones, the platform can get the
bearing estimation of each array by running AML algorithm
and location estimation by running LS algorithm. The design
and implement of this platform is simple but it provides a
general frame for target localization based on acoustic array
sensor networks. UCLA has also developed a microphone
array network by using of iPAQ3760s which achieves the
localization of acoustic target [2]. In this wok, practical Figure 2. Multi-layer system structure
coherent array processing issues are considered, including the
propagation noises and time synchronization. This platform
builds a totally distributed sensor network structure within A. Hardware Layer
which virtual array substitutes traditional array and it provides The hardware of node consists of three parts: acoustic array,
new direction for WSN on tracking. In an integrated sensing power supply module and processing module. Acoustic array is
platform done by UCB achieves multi-target tracking by taking made up of 4-channel or 6-channel microphones laid out in an
advantages of co-processing acoustic, vibration, visual and circle and the azimuth angle is defined in a uniform regulation.
other signals [1]. The well-designed system architecture makes The node is powered by a Li rechargeable battery. Regulator
it possible for different sensors to work cooperatively. Also, in circuit and voltage conversion circuit are designed to provide
military applications, many projects have been done such as stable +/-12V voltage for sampling chip and +/-5V voltage for
gunfire localization system [6] and intrusion detection system other chips.
[7] based on acoustic monitoring. U.S. Army Research
Laboratory has done lots of work in this field. A robot-based The processing module, shown as Figure 3, is based on
acoustic detection system was developed to detect and localize MSP430, FPGA and DSP. FPGA as the master device mainly
on impulsive noise events [8]. Along with some helmet- perform the following tasks: 1) producing the timing of the
mounted acoustic array labeled by soldiers, the whole system signal sample (AD); 2) controlling the wireless module to
can produce an accurate location of a target. complete the transmission of information; 3) controlling the
wireless module and self-localization module to complete
As a summary, acoustic target sensing platforms have been synchronization and self-localization; 4) providing some
developed for target detection, localization and tracking. Most necessary expansion interface, to facilitate the function of the
of the previous work is for single-target and real time tracking system expansion and system upgrading. As the core device of
service can be hardly provided. IAASNT, however, is designed signal processing, the main function of DSP is to run the core
for on-time multi-target tracking. Also, it provides an algorithm and provide application service, such as target
integrated tracking service which has rarely been considered in tracking and localization. The communication between FPGA
acoustic array sensor networks in a system view. and DSP is based on the data bus and address bus. Some
memory devices, such as SDRAM and Flash are also mounted

324
on the bus to be used in the intermediate results storing. Time synchronization is necessary for application platforms.
MSP430 is the MCU in the IAASNT system which achieves In order to synchronize nodes, we design a lightweight scheme
low-power work mode. as following. All nodes are synchronized by sink node. The
sink node broadcasts a synchronization message with MAC-
Wireless
module
AD layer timestamp and seqNum every 10 seconds. The nodes in
the broadcast radius of sink node collects reference point and
then broadcast a synchronization message. By this way, all the
FRQWURO nodes can collect reference point directly from the sink node or
MSP430 Add(0:15) FPGA Add(0:15) DSP
indirectly. By this scheme, the network can be synchronized
FIFO Data(0:31)
from sink node to the normal nodes within 1ms difference.
Data(0:31)

Node Localization is also an important component for

Data˄0:31˅
SDRAM
tracking. Different from the traditional node localization, node
Add˄0:13˅ localization in IAASNT needs to decide not only the position
SpiderBat Interface
Data˄0:7˅
but also the orientation of node. A special mechanism is
Add˄0:18˅
Flash designed to accomplish the task [11]. During processing, a
moving media-object passes through the monitoring area so
Figure 3. Structure of node processing module that each node can sense the media-object and report its
bearing measurements. In the meantime, the position of media-
object is obtained by GPS. By using bearing measurements of
B. Middleware Layer
nodes and the position of media-object, node localization can
The components in middleware layer are the link between be easily achieved by maximum likelihood algorithm and the
hardware layer and application layer. To provide real-time cost function can be written as:
bearing measurements for tracking service, signal process,
2
communication and synchronization are necessary and are
1 §N py − y ·
achieved in node. Node localization is executed when system is Cost = ¦ 2 ¨ θi − arctan i +β¸ (1)
deployed and is done in fusion center. ¨
i =1 σ i © pxi − x ¸
¹
Signal process is mainly responsible for accomplishing
Direction of Arrival (DOA) estimation. DOA estimation has Here, pxi , p yi is the position of media-object and x, y is
been studied for decades and in our system, a method named
Focusing Khatri-Rao subspace method (FKR) [9] is proposed the position of node to be localized. θi is the bearing
based on coherent signal-subspace method (CSM) and Khatri- measurement of node, and β is the orientation which is also
Rao (KR) subspace. FKR includes two primary steps: focusing
and arrangement. After focusing, the DOAs are estimated an unknown parameter to be determined. σ i2 is the variance of
according to the property of KR product. measurement error.
For simplify, communication follows the basic protocols
for WSN but we make a modification into the MAC layer C. Application Layer
packet structure, named Q-MAC [10], to achieve adjustable The application layer concerns with data processing on
bandwidth communication. In Q-MAC, superframe structure target tracking and an integrated tracking frame is designed as
and support multi-hop packet transmission are adopted, see shown in Figure 5. The frame designed for IAASN includes
Figure 4. The first byte of the packet payload is a load indicator modules such as tracking filtering, data association, tracking
variable which indicates the number of current queued packets initiation similar to the classical tracking frame, but also
in the node’s MAC layer. Receiving the data packets, thus includes modules such as node selection, initial state estimation
receiving the indicator variable, the cluster head knows the designed since multi-sensor and bearing-only measurement
senders’ packet load by extracting the indicator variable from bring in new problems.
the data packet. It then accordingly allocates certain TDMA Tracking Modules
slots into the next superframe period to compensate the queued
packets of the son-devices. To give a chance of knowing Initial state Track initiation
estimation (termination)
scattered traffic loads of all son-devices, a fixed length CSMA
period follows the variable TDMA period. Input Output

Bearing Data Tracking

Target state
measurement association filtering

Node selection
Schedule Instruct

Figure 5. Integrated target tracking frame

Figure 4. Superframe sturcture
The classical modules for tracking follow the common
algorithms such as interacting multiple model (IMM) algorithm
[12] for filtering, joint probabilistic data association (JPDA)

325
[13] for data association. For the limitation of the paper size, iteration, as shown in algorithm 2. In initial state estimation
the details are omitted and we emphasize on introducing node part, a probability is calculated while localization to give a
selection and initial state estimation parts. relative division of targets and ghosts which is decided by the
residual between the predicted azimuths and the bearing
Node selection for target tracking consists of two main measures. Then, by setting a distance taboo, only a few
parts: establishing a cost function for weighing localization relatively independent localization outputs are selected and
accuracy and optimizing the cost function to obtain node new tracks to be initiated are created based on the localizations
selection strategy. Node selection under bearing only and their probabilities. The reason why we choose only one
measurements is quite different from common node selection scan to estimate initial state is that measurements from targets
in WSN for the cost function is quite difficult in this case. The may be not stable and localization results are not accurate
geometrical dilution of precision (GDOP) [14] is a fine cost enough so it is hard to estimate a valid target velocity by using
function in theory and many related works have been done multiple localization from multiple scans. In order to avoid the
based on it. The difficulty mainly lies on the searching process influence on lacking of initial velocity estimation, covariance
to optimize the cost function which is a NP-hard problem. of target motion noise should be set larger at first. In track
Instead of complex process for an optimal or suboptimal node initiation part, probability-based algorithm similar to [16][17]
selection result, IAASNT uses a simple heuristic approach is adopted. Probability of target existence updates while
similar to the “add one node at a time” method [15] to achieve iteration and tracking initiation strategy can be made based on
an acceptable result, see Algorithm 1. Firstly, neighboring it.
nodes are selected as candidates. Then, by finding the
minimizing cost function f cost proposed in [15], the first Algorithm 2 Initial state estimation and track initiation
selected node into Ns is obtained. By finding one node Input: measurements Z = {zi ( j )} received from each node, record for
minimizing cost function each time, node selection process can
track to be confirmed TraIni = {TI i }
be accomplished. Although it is not an optimal approach, the
method can achieve a relative fine result in a short time which Output: record for track confirmed TraCon = {TCi }
meets the requirement for on-line process.
1: Loc = ∅ , PLoc = ∅ ;
Algorithm 1 Node Selection: heuristic approach 2: for each group consists of 3 neighboring sensors do
Na , the predicted target position Xt
Input: the position of all the nodes 3: [ Loc , PLoc ]=MultiLocalization( Z );
and covariance Pt , the expected number of node selection num( Ns ) 4: end for

Output: node selection output Ns 5: do LocMerge on [ Loc , PLoc ] ;

1: find min( 2 num( Ns ) , num( Na ) ) neighboring nodes 6: create new track TI i to TraIni by [ Loc , PLoc ];
{Nri :1 ≤ i ≤ num( Nr )} to predicted target position from Na ; 7: for each TI i from TraIni do

2: for each Nri , i = 1,..., num( Nr ) do 8: TI i ,k +1 =TrackIteration( TI i ,k );

3: calculate f cos t ,i ( Nri , Xt , Pt ) ; 9: if TI i ,k +1 → PTarPro ≤ PTarTer then
4: end for
5. selectNrk from Nr for minimize f cos t as the first node of Ns ; 10: delete TI i from TraIni ;
else if TI i , k +1 → PTarPro ≥ PTarCon
6: for j = 2,..., num( Ns ) do
11: or supplementary condition
then
7: find the minimize f cos t by Ns and Nrj from each Nri not in
12: add TI i to TraCon , delete TI i from TraIni ;
Ns ; 13: end if
8: Ns = Ns ∪ Nrj ; f cos t = f cos t ,min ; 14: end for

9: end for
D. Low-power Management
Low-power management technique is a necessary part for
Initial state estimation is needed on tracking initiation for an applied system especially for wireless sensor network
under bearing only measurements the initial state is not ready system. The main idea of low-power management is
for us. The problem of initial state estimation is in fact multi- scheduling nodes to work in active mode only when necessary.
target localization under bearing only measurements. Single In order to achieve reliable service with low-power consuming,
target localization under bearing measurements is not difficult a multi-level low-power management is proposed for IAASNT
while multi-target localization algorithm is rare. In fact, it is as shown in Figure 6. The management can be divided into
almost impossible to absolutely distinguish target and ghost in three levels corresponding to the multi-layer structure of
only one scan under measurement suffering undetected and system and we will describe them in bottom-up flow.
false-alarm. We design an integrated mechanism in which
initial state estimation part is responsible for providing a In hardware level, the design and implement of sensor node
probability based initial state estimation but leaves more satisfies the basic regulations for low-power consuming. Also,
precise work to the target initiation part during several scans’ the hardware structure makes it possible for the node to change

326
its mode according to decisions made by upper levels to level are made to reduce the redundancy both in time and space
achieve energy saving. in order to achieve more energy saving. The scheduling is
based on the result of node selection module in target tracking
System Decision Application
component.
Target-finding Stage Target-tracking Stage
Node Decision
Middleware Components *URXS'HWHFWLRQ 7UDFN,QLWLDWLRQ 7UDFN7UDFNLQJ
Middleware
Signal Network
Process Protocol
Fail
1RGH'HWHFWLRQ 1RGH6HOHFWLRQ

Low-power Active
Mode Mode
Figure 7. Low-power management in system level
Hardware
Hardware Structure

IV. EXPERIMENTS AND ANALYSIS

Figure 6. Multi-level low-power management frame We have performed a number of experiments to evaluate
tracking service of IAASN in two environments: single target
In middleware level, low-power mode is designed for tracking in noise-free environment and multi-target tracking in
signal process component and also mode transition strategy is noisy environment.
proposed. For signal processing, the calculation for DOA
estimation needs to be executed by DSP which is high power A. Noise-Free Environment Experiment
consumption relatively. So low-power mode is designed with
DSP shut down to save energy and only low power consuming The noise-free experiment [18] is performed in an open
MSP430 is working to make decisions for mode-transition. The area more than 1000*1000m. Five sensor nodes are randomly
comparison of power consumption for the two modes is as disposed on the flat glass ground about 300*300m, see Figure
shown in Table 1. The decision is made by a single node 8. The real position of each node is obtained by GPS device. In
without any knowledge of other nodes. In detail, by calculating order to test node localization component, an assistant object
the correlation coefficient of the data in frequency domain from with GPS moves through the area and it is easy to localize each
any two array elements, a relatively credible conclusion can be node. Figure 9 shows the result of node localization comparing
made on whether the target exists or not. In order to keep a with the real position.
relatively stable detection, a mechanism similar to Schmitt
Trigger is utilized to avoid too many transitions which will lead
to heavy system cost for scheduling and confusion on system.
Also, the proposed low-duty network protocol can work well
under different modes and achieves further energy saving.

TABLE I. POWER CONSUMPTION FOR TWO MODES

Mode\device DSP FPGA MSP430 others total

Active 1.23w 0.31w 0.05w 0.78w 2.37w
Figure 8. Field for experiment in Figure 9. Nodes deployment and
Low-power 0 0.31w 0.05w 0.2w 0.56w
noise-free environment localization

In application level, the process for service can be divided 150 20

into two stages, these are target-finding stage and target- 100
Node
GPS: target true loca
target estimate track
18

16
tracking stage. Strategies are designed for the two stages 14
50
respectively to make system decision for scheduling nodes, see 12
error:m
Y:m

Figure 7. In target-finding stage, due to limitation of node 0 10

sensing and lacking of a global view for tracking service, -50

6

decisions made in node level may be not reliable enough, -100

4

which will lead to delay or mistake in target finding. So the -150

-150 -100 -50 0 50 100 150
0
0 20 40 60 80 100 120 140 160 180 200

main challenge lies on problems of dealing with node X:m time:s

undetected. A group neighboring cooperation strategy is (a) (b)

proposed to solve the problem. The node detection decisions Figure 10. Comparison of estimate and real track: (a) track plot (b) error
are reported to fusion center and detection nodes are grouped
according to their positions. If the number of nodes in group is For testing performance of target tracking, we put a GPS
more than 2, nothing else needs to do. Otherwise, neighboring device on target and get the ideal target position in real time.
nodes of the group will be set active so that the number of The target moves around each node and we get the estimated
active nodes in group reaches 3 which is necessary for initial target track and the real track in the same time, see Figure 10(a):
state estimation in tracking. While in target-tracking stage, the red point denotes the ideal target position obtained by GPS
information from nodes is redundant and strategies in system and the blue point is the estimated track obtained by this

327
system. The error analysis is shown as Figure 10(b). The Besides algorithms mentioned above, the design of low-cost
average error is about 5.8m. Considering that the target is about and low-power sensor is a great challenge. Also, theoretical
5m long, the tracking performance is perfect. and experimental study for problems under large-scale of
sensors is not yet sufficient which will be the future work.
B. Noisy Environment Experiment
In application, more complex case must be considered. To ACKNOWLEDGMENT
test the performance, we do experiment in a noisy environment The authors would like to thank Ji-an Luo, Kai Yu and
[19]. In this experiment, the field available is only about Ming Yin for their contributions.
300*300m and the sensors are deployed relatively closely.
In this experiment, we do multi-target tracking test. Two REFERENCES
vehicles are moving in the experiment field at the same time. [1] Oh, S., L. Schenato, et al. “Tracking and coordination of multiple
Figure 11 (a) shows the tracking result (blue point) comparing agents using sensor networks: system design, algorithms and
with real position (red point) and (b) is the error analysis. The experiments.” Proceedings of the IEEE. 2007, 95(1):234-254.
average errors for two targets are about 8.1m and 7.3m [2] Chen, J. C., L. Yip, et al. “Coherent acoustic array processing and
respectively. Figure 12 shows the initial state estimation when localization on wireless sensor networks.” Proceedings of the IEEE.
target starts moving and in this time four potential tracks are 2003, 91(8): 1154-1162.
ready for initiation. Figure 13 is the probabilities of target [3] Girod L, Lukac M, Trifa V, et al. “The design and implementation of a
self-calibrating distributed acoustic sensing platform.” In ACM
existence for the two tracks which have been successfully Proceedings of the 4th international conference on Embedded networked
initiated. Track 1# (blue line) can be quickly initiated for the sensor systems. 2006: 71-84.
probability of the track exceeds initiation threshold within 3s. [4] Allen M, Girod L, Newton R, et al. “Voxnet: An interactive, rapid
While the probability of Track 2# (green line) increases slowly, deployable acoustic monitoring platform.” In IPSN ’08: Proceedings of
under the help of supplement strategy, it is also initiated within the 7th international conference on Information processing in sensor
6-7s. Also, it can tolerant bad measurements and even networks. St. Louis, Missouri, 2008.
undetected in some degree, see Track 2# at about 105s. [5] Wang, H., C. E. Chen, et al. “Acoustic sensor networks for woodpecker
localization.” Proc. of SPIE, 2005
150
sensor
30
Error for 1#Target
[6] Ledeczi A, et al. “Countersniper system for urban warfare.” ACM
100
1#target
2#target
estimation
estimation 25
Error for 2#Target
Transactions on Sensor Networks. 2005(2):153–177.
1#target real postion

50
2#target real postion
20
[7] Arora A, Ramnath R, Ertin E, et al. “Exscal: Elements of an extreme
2
4
scale wireless sensor network.” In IEEE Proceedings. 11th IEEE
error:m

5
International Conference. 2005:102-108.
Y/m

0 15

7
-50 1 10 [8] Young, S. H. “Detection and localization with an acoustic array on a
-100 5
small robotic platform in urban environments.” Army Research Lab
Adelphi MD. 2003.
-150 0
-150 -100 -50 0
X/m
50 100 150 20 40 60 80
time:s
100 120 140
[9] Feng, D, M. Bao, et al. “A Novel Wideband DOA Estimator based on
Khatri-Rao Subspace Approach.” Signal Processing. 2011:0165-1684.
(a) (b)
[10] Zhuo S.G, Song Y.Q, Wang Z. Wang Z.B. “Queue-MAC: A queue-
Figure 11. Comparison of estimate and real track: (a) track plot (b) error length aware hybrid CSMA/TDMA MAC protocol for providing
dynamic adaptation to traffic and duty-cycle variation in wireless sensor
150 1 networks”, 9th IEEE International Workshop on Factory
nodes
localization
0.9 Communication Systems (WFCS2012), May 21-24, 2012,
100
0.8

0.7
[11] Xue H.L, Wang Z. Luo J.A. “Self-localization Mechanisms for Acoustic
50
4
2
0.6
Array Network Based on Mobile Beacon and DOA.” Information and
probability

5
control, 2011, 40(6).
Y:m

0 0.5

0.4
-50
7
1
0.3
[12] Blom, H. A. P., and Bar-Shalom, Y. “The interacting multiple model
0.2 1#Track
algorithm for systems with Markovian switching coefficients.” IEEE
-100
0.1
2#Track
Transactions on Automatic Control, 33, 8 (Aug. 1988), 780—783.
-150
-150 -100 -50 0 50 100 150
0
20 40 60 80
time• s
100 120 140 [13] Fortmann T.E., Bar-Shalom Y. and Scheffe M. “Multi-target using joint
X:m
probabilistic data association.” In Proceedings of the 19th IEEE
Figure 12. Initial state estimation for Figure 13. Calculation for probability of Conference on Decision and Control,(2),1980: 807-812.
track initiation track [14] Kadar, I. “Optimum geometry selection for sensor fusion.” In
Proceedings of SPIE, 3374 1998: 96-107.
[15] Kaplan, L. M. “Global node selection for target localization in a
V. CONCLUSION distributed network of bearings-only sensors.” IEEE Transactions on
Aerospace and Electronic Systems, 2006,42(1): 113-135.
Research in WSN for tracking has been active in the past
[16] Musicki, D., R. Evans, et al. “Integrated probabilistic data association.”,
ten years. Most of the works faces difficulties on real-time IEEE Transactions onAutomatic Control, 1994, 39(6): 1237-1241.
processing and multi-target problems. While in IAASNT, the
[17] Li, N. and X. R. Li. “Target perceivability and its applications.”, IEEE
designing of system structure and the implement of each Transactions on Signal Processing, 2001, 49(11): 2588-2604.
component makes it possible to provide an integrated on-line [18] Luo, J., D. Feng, et al. “Experiments for on-line bearing-only target
multi-target tracking service which can rarely be seen at localization in acoustic array sensor networks.” 8th World Congress on
present. It is a significant work from other similar systems. Intelligent Control and Automation (WCICA), 2010: 1425-1428.
Also, the experiments show a fine performance of the system. [19] Li Y.S, Wang Z., Bao M. etc. “The design and experiments for multi-
However, there are still problems to be solved in the future. target tracking platform based on acoustic array sensor networks.”
Journal of Scientific Instrument. 2012,33(1):146-154

328