You are on page 1of 13

IEEE TRANSACTIONS ON SYSTEMS, MAN, AND CYBERNETICS: SYSTEMS VOLUME: 47 NO.

2 FEB 2017
1

Preprocessing Design in Pyroelectric Infrared


Sensor-Based Human-Tracking System:
On Sensor Selection and Calibration
Jiang Lu, Member, IEEE, Ting Zhang, Fei Hu, and Qi Hao

AbstractThis paper presents an information-gain-based sen- tracking performance. There are two main preprocessing steps:
sor selection approach as well as a sensor sensing probability 1) sensor selection and 2) calibration.
model-based calibration process for multihuman tracking in Sensor selection is an important problem in human track-
distributed binary pyroelectric infrared sensor networks. This
research includes three contributions: 1) choose the subset of ing based on distributed sensor networks. As sensors become
sensors that can maximize the mutual information between sen- ubiquitous, coupled with the increasing demand for a shorter
sors and targets; 2) find the sensor sensing probability model to processing time and less resources, it becomes important to
represent the sensing space for sensor calibration; and 3) pro- perform selective fusion so that the decision can be made in
vide a factor graph-based message passing scheme for distributed a timely and efficient manner. The need for sensor selection
tracking. Our approach can find the solution for sensor selection
to optimize the performance of tracking. The sensing probability is further demonstrated by the burden of an increasingly large
model is efficiently optimized through the calibration process in volume of sensory data and by the variability of sensor oper-
order to update the parameters of sensor positions and rotations. ations over time and location. We can reduce the amount of
An application for mobile calibration and tracking is developed. data and improve fusion accuracy by selecting proper sensors
Simulation and experimental results are provided to validate the in a certain location at a certain time.
proposed framework.
Calibration is another important issue in distributed track-
Index TermsBinary sensor networks, calibration, factor ing sensor networks. The tracking algorithms need to read
graph, information gain, matrix factorization, message passing, the geometric parameters of sensors: position and orientation.
multiple human tracking, sensor selection.
The estimation errors of sensor parameters result in degraded
tracking performance. This problem is even worse in multi-
ple target tracking, where the errors of one tracker propagate
into other trackers through the data-to-target association pro-
I. I NTRODUCTION cess [7]. Therefore, a successful distributed tracking system
NDOOR multihuman tracking has many applications requires a calibration component that can estimate the sensor
I in surveillance, security, healthcare, and energy-efficient
buildings [1][4]. Recently, binary sensing technologies have
geometric parameters accurately.
This paper presents a framework for distributed tracking
enabled the development of a distributed binary sensor system with dynamic sensor selection and calibration using binary
for target tracking and recognition [5], [6]. The advantages of pyroelectric infrared (PIR) sensors. The contributions of this
a distributed binary tracking system include: 1) system scal- paper include two aspects.
ability to a large number of sensors; 2) system robustness to 1) Information-Gain-Based Sensor Selection: We proposed
local failures; and 3) a system that is low cost with small an information-gain-based sensor selection method for
sensor communication overhead. selecting sensors that can maximize the mutual infor-
In a complete sensor-based tracking system, typically, pre- mation between sensors and targets. The multiple target
processing is needed before tracking in order to achieve higher information gain is derived and formulated. Given the
localizations of targets, we prove that subregion sensor
selection is better than the whole-space sensor selection
Manuscript received April 7, 2015; revised July 27, 2015; accepted by using less sensors with similar tracking errors.
November 2, 2015. This paper was recommended by Associate 2) Sensing Probability Model for Calibration: The calibra-
Editor M.-Y. Lee. tion method updates the probability model by using the
J. Lu is with the Department of Computer Engineering, University of
Houston at Clear Lake, Houston, TX 77058 USA (e-mail: luj@uhcl.edu). sparse measurements. A mobile application is developed
T. Zhang and F. Hu are with the Department of Electrical and Computer to customize calibration paths. It allows the users to do
Engineering, University of Alabama, Tuscaloosa, AL 35487 USA (e-mail: calibration without covering the entire sampling space.
tzhang21@crimson.ua.edu; fei@eng.ua.edu).
Q. Hao is with the Department of Electrical Engineering, South University The rest of this paper is organized as follows. Section II
of Science and Technology of China, Shenzhen, Guangdong 518055, China reviews the related work. Section III provides the big picture
(e-mail: hao.q@sustc.edu.cn). of our preprocessing method as well as the system model.
Color versions of one or more of the figures in this paper are available
online at http://ieeexplore.ieee.org. Section IV presents an information-gain-based sensor selec-
Digital Object Identifier 10.1109/TSMC.2016.2523914 tion method. Section V provides the sensing probability model
2168-2216 c 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

2 IEEE TRANSACTIONS ON SYSTEMS, MAN, AND CYBERNETICS: SYSTEMS

and factor analysis-based calibration. In Section VI, we give sensor parameters. In this paper, we use the model-based
the factor graph model-based distributed belief propagation calibration. The probability of sensor sensing model can
framework for multiple target tracking. Section VII provides be updated by gradient descent algorithm. Then, the senor
the simulation and experimental results. Finally, Section VIII parameters can be computed from the updated sensor sensing
concludes this paper. probability model.
The above preprocessing tasks (sensor selection and cali-
bration) prepare us for PIR sensor-based multihuman tracking.
II. R ELATED W ORK We will briefly introduce our tracking scheme here to illus-
Multihuman tracking and recognition have been devel- trate the effects such as reducing tracking errors after sensor
oped for wireless PIR sensor networks [6], [8], [9]. selection and calibration. We will use factor graph to model
Furthermore, human activity learning using context is also our tracking process since factor graph can well represent the
studied [10], [11]. However, no previous studies in PIR sensor message passing process among distributed PIR sensors.
networks clearly present the sensor selection scheme to give Factor graphs are a straightforward generalization of the
the optimal solution. Tanner graph of Wiberg et al. [27], [28]. Tanner [29] intro-
The existing methods for sensor selection can be gener- duced the bipartite graphs to describe families of codes which
ally divided into two groups: 1) the Bayesian-based methods are generalizations of the low-density parity-check (LDPC)
and 2) the information-based methods. The Bayesian-based codes of Gallager [30] and also described the sum-product
sensor selection considers the whole process as an optimiza- algorithm in this setting. Factor graphs take these graph-
tion problem to find the best solution among all possible theoretic models one step further, by applying them to func-
sensor combination. In [12], a Bayesian I-optimal approach tions. From the factor-graph perspective, a Tanner graph for
is used. Giraud and Jouvencel [13] proposed a geometrical a code represents a particular factorization of the characteris-
approach for the sensor selection. Garnett et al. [14] con- tic function of the code. Factor-graph-based tracking has been
structed covariance functions and performed Gaussian process used in many applications. In inertial navigation systems, the
inference over the functions whose domains are a power set. factor graph has been used for information fusion [31]. A mul-
Ni and Pottie [15] used a Bayesian detection approach to tiple hypothesis tracking with factor graph has been proposed
select a subset of sensors. These methods define the selec- in [32]. The factor graph-based association scheme over joint
tion process as probability problems. The sensor selection probabilistic data association has been developed for closely
can be represented as conditional probability given the sen- moving targets in [33]. In this paper, the factor graph is an
sor observations and background information. Most methods undirected graph which is efficiently used in message passing
of information-based sensor selection in tracking problems algorithms.
compute the distribution of target locations and then select
sensors which give the minimum entropy of the target loca-
III. S YSTEM M ODEL
tion distribution [16][18]. In [19], the KullbackLeibler and
Chernoff sensor selection criteria is provided for event detec- A. Binary Sensor System
tion in wireless sensor network. Zhang and Ji [20] presented The low-cost, low-power PIR sensors are used in our
an interesting approach to calculate information gain between binary sensor system. Simple binary bits are generated, which
two sensors. In our research, the information-based method represent the sensors as triggered (1) or not (0), as
is chosen for sensor selection. We only use the binary codes shown in Fig. 1(a). The advantages of binary sensing include:
generated by sensors to estimate the information gain before 1) tolerance of low signal-to-noise ratios and 2) reduction
and after adding sensors. Our method also considers the tar- of data throughput and communication overhead. We have
get locations as prior knowledge for sensor selection to narrow developed pyroelectric sensor nodes for human tracking and
down the coding space. recognition. Each sensor node contains two PIR sensors.
In most sensor networks, the objective functions of Fresnel lens arrays and masks are used to change the field of
the calibration can be constructed through two approaches: view (FOV) of each sensor. As illustrated in Fig. 1(b), the FOV
1) the model-based method or 2) the data-driven method. In the of one sensor node is segmented into many small partitions.
model-based framework, measures of sensor readings can be Each partition is associated with a sequence of binary code,
predicted based on the sensor models. The sensor parameters e.g., 10, 01, and 11. When human subjects walk through
are then self-tuned to reduce the prediction errors [21][23]. the FOV, the sensor node can generate the binary sequence
In the data-driven framework, measures of sensor readings that can be analyzed to find the location, gesture and activity
are correlated with regression functions of sensor parameters. information of targets.
A two-stage process is usually involved: 1) identification of Now, let the M sensors, S = S1 , S2 , . . . , SM , segment the
regression function parameters and 2) fine-tuning of sensor tracking space into many small regions. Each region is asso-
parameters [24][26]. ciated with an M-bit binary code. Fig. 2(a) shows an example
The advantages of data-driven methods include their inde- of space encoding by three distributed binary PIR sensors.
pendence of sensor physical models and the maturity of With such a space encoding scheme, each code measurement
related algorithms. However, those methods usually assume is associated with one or multiple possible regions. Therefore,
slow varying stimulus sources and the existence of well con- a space decoding scheme has to be developed. In [6], we have
ditioned regression functions between sensor readings and used data-to-target association methods for space decoding.
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

LU et al.: PREPROCESSING DESIGN IN PIR SENSOR-BASED HUMAN-TRACKING SYSTEM 3

the information gain by given previous subset U and current


adding sensor Si . Mathematically, the conditional information
gain (CIG) is defined as
     
J(X, U) I X; Si |U = H X|U H X|U , Si (2)

where Si is selected from U and U = U Si .


We still wish to maximize the dynamic CIG to find the
Fig. 1. Illustration of (a) binary multiplex sensing and (b) two coded masks sensor Si which provides the maximum information to min-
for PIR sensors. imize the tracking error of the system. When the number of
sensors m is not given, the dynamic sensor selection process
stops adding sensors when the CIG is less than a threshold

J(X, U) (3)

where is a small value less than 1.

C. Calibration
Let the whole system sampling geometry be G(, ), where
Fig. 2. Illustration of (a) space encoding by three distributed binary PIR
G(, ) = g1 (1 , 1 ), g2 (2 , 2 ), . . . , gm (m , m ). It is deter-
sensors and (b) ideal gridded space encoding case with one target tracking. mined by the positions and orientations of the sensors.
With such a sampling geometry, we can achieve tracking for
multiple targets [6].
The measurement decoding converts codewords into the loca- The goal of calibration is to find these parameters that can
tions. This space encoding/decoding theme is used for target perform the best tracking results, i.e., minimizing the track-
tracking. ing errors. In this paper, we propose a new way to achieve
Assume the tracking space is segmented into N small calibration: instead of minimizing the tracking errors between
regions. Then, the binary codes for these N grids are C = the measured and predicted positions, we minimize the error
c1 , c2 , . . . , cN . ci is an M 1 binary string. An approximate between the measured and predicted sampling geometries.
way to represent these small regions are grids. A simple ideal Let the measured sensing probability matrix be V. It can be
example of gridded space encoding with one target tracking approximately represented by two nonnegative matrices, the
by M = 4 sensors is given in Fig. 2(b). This gridded ideal predicted sensing probability basis matrix G RRT + and
space encoding method is used in sensor calibration. weighting matrix W RFR + , that is

V V = WG (4)
B. Sensor Selection
Presence of multiple targets will cause more ambiguities for where the measured sampling probability matrix V(, ) can
such a binary sensor system. In order to optimize the sensor be calculated from G(, ) and the number of basis in G can
deployment to achieve accurate tracking, sensor selection has be given by R < (FT/(F + T)). The initial predicted posi-
to be made. On one hand, multiple sensors have to be triggered tions and orientations of sensors are known to be prior
to track the whole trajectory; on the other hand, we should knowledge.
not trigger too many sensors to avoid high communication The cost function of calibration for the sensor i, in Euclidean
overhead and sensor energy waste. distance form, can be written as
Assume that there are M binary sensors in the sensor net-
work, and the whole tracking space is segmented into N Ei (i , i ) =
Vi (i , i ) Wi Gi
2 +
Wi
+
Gi
(5)
small grids. Also, assume that the states of K targets are
X = (X 1 , X 2 , . . . , X K ). The state can contain the location and where

is L1 norm, and and are the coefficients of the


velocity information of the target. The objective of the sen- penalty terms of discriminant information. We attempt to find
sor selection is to select a subset U of m sensors, U S, to the Vi = Wi Gi with optimal parameters i ,i that can
maximize the mutual information of X and U arg min Ei (i , i ) s.t. Wi 0, Gi 0. (6)

i ,i
arg max I(X; U) (1)
US
The whole calibration process is illustrated in Fig. 3. The
where I(X; U) = H(X) H(X|U), and H(X|U) is the predicted sensing probability basis matrix G will be updated
conditional entropy given the subset U. by giving the measurement sensing probability matrix V. Then
However, selecting the subset U at one time is difficult. the calibration parameters can be computed from the optimized
We have to explicitly consider all the possible combinations reconstructed sensing probability matrix V . In this process,
from S, which will be computationally inefficient. Thus, we the parameters of each sensor will be calibrated separately.
provide a dynamic sensor selection theme which computes The details will be given in Section V.
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

4 IEEE TRANSACTIONS ON SYSTEMS, MAN, AND CYBERNETICS: SYSTEMS

Fig. 3. Illustration of matrix factorization-based calibration process.


Fig. 5. System diagram of distributed multiple target tracking with sensor
selection and calibration.

not all of them provide useful information that improves the


estimate. Sensor selection aims to select the optimal subset of
sensors that give the best tracking performance.

Fig. 4. Illustration of the sum-product update rule for factor graph. A. Sensor Node Information Measure
When the sensors are deployed in the distributed sensor
network, the observation space is segmented into many small
D. Message Passing
regions similar as shown in Fig. 2(a). Assume the number of
Because our tracking algorithm is based on message pass- regions is N. Generally, m binary sensors in subset U mean that
ing, here we briefly explain its principle. Our message passing each region can be represented as an m-bit binary code. There
algorithm uses a factor graph. Let f x denote the message are 2m 1 codes (without zeroes), Cm = (cm 1 , c2 , . . . , c2m 1 ),
m m
sent from fusion node f to variable node x. Let xf denote p
where cq denotes the qth binary code which is generated by p
the message sent from variable node x to fusion node f . Also, binary sensors. Each code has the probability of detecting one
let ne(x) denote the set of neighbors of a given node x in the target as
factor graph. As shown in Fig. 4, the sum-product message    
1
passing algorithm can be illustrated as two expressions i = m,
p Xj |cm i 1, . . . , 2m 1 (9)
 ni
xf (x) = uj x (x) (7) where nm m
i is the number of regions associated with code ci ,
uj ne(x)\ f m
and Xj denotes the jth location with code ci . Then, the
  
f x (x) = f (v1 , . . . , vM ) vi f (vi ) . (8) information of locating one target given m binary sensors is
v1 vM vi ne( f )\x 1
2 m
     
HU X|Cm = i H X|C = ci
p cm m m
The factor graph-based message passing algorithm for target
i=1
tracking will be discussed in details in Section VI. m 1
2 nm
 i
   
= p cm
i p Xj |cm
i log2 p Xj |ci .
m
(10)
E. System Diagram
i=1 j=1
The whole system diagram is illustrated in Fig. 5. By select-
Here, we denote HU (X|Cm ) = H(X|U).
ing the sensor set U from sensors S, we have the knowledge
For the case of multiple targets tracking, the probability of K
of sampling geometry G(, ). Such a set of sensors can gen-
targets locating in K regions with one of (2m 1) possible
erate the codes for the tracking space. When the targets are
codes is given by
walking through the encoded space, a binary event sequence is
generated. A calibration process will be used to update sensing     K 
 
probability models. In the meantime, the distributed message p X 1,...,K cm
i = p X k cm
i (11)
passing algorithm is performed for multiple target tracking. k=1

The target states will be updated during the message passing where we assume K nm i . As the tracking processes
procedure. for all targets are independent, we can rewrite (11) as
In the next few sections, we will provide the details for each p(X 1,...,K |cm
i ) = p(X |ci ) . The conditional entropy for K
1 m K

module of Fig. 5. targets given subset U is defined as


     
1) Formulate the information-gain-based sensor selection H X 1,...,K U = HU X 1,...,K Cm
problem via information gain in PIR sensor networks.
nm K

   
m 1
2
2) Provide the sensing probability model for calibration,  m i
k m
and use nonnegative matrix factorization algorithm to = p c p X c i j i
update the model. i=1 j=1 k=1
K 

3) Build a factor graph-based distributed message passing   


k m
model for multiple target tracking. log2 p Xj ci
k=1
m
IV. I NFORMATION -G AIN -BASED S ENSOR S ELECTION m 1
2 ni
 m    K
Given the distributed binary sensor network, we wish to find = K p ci p Xj1 cm
i
i=1 j=1
the minimum subset of sensors that can provide the best track-   
ing estimation. Among all available sensors in the network, log2 p Xj1 cm
i . (12)
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

LU et al.: PREPROCESSING DESIGN IN PIR SENSOR-BASED HUMAN-TRACKING SYSTEM 5

Fig. 7. Illustration of sensing probability model of calibration.

Fig. 6. Illustration of sensor selection process. C. Subregion Sensor Selection


The CIG is related to target location and sampling geom-
Please note that 2m 1 are the maximum possible codes etry. So far, the CIG for sensor selection is computed based
that m sensors can generate. The actual number of different on the whole tracking space. However, it is not necessary to
codes is less than or equal to 2m 1. maximize the CIG in the whole space in many cases. For
example, for one target case, it is more efficient to maximize
B. Conditional Information Gain the CIG in a subregion space based on the tracking estimation
As given in (2), the objective function of CIG is the gain of of the target location. In another case, if we have the prior
conditional entropy between the current subset U and previous knowledge of target trajectories, we can maximize the CIGs
subset U . We aim to find a sensor Si , which is not included of several subregions. That is, when we select sensors, we use
in subset U , to maximize the CIG I(X; Si |U ). The sensor the subregions in which targets will appear more likely than
selection process will terminate when the maximum CIG is others to maximize the CIG.
smaller than a threshold . In other words, the added sensor
does not provide much information for more accurate tracking
V. M ATRIX FACTORIZATION -BASED C ALIBRATION
in the network. The process terminates when the number of
selection steps reaches the desired number. The process of CIG A. Sensing Probability Model
is shown in Fig. 6. At each step, one sensor is selected based As we described in Section III-A, the sensors segment the
on the maximum CIG. space into multiple small regions. Fig. 7 shows the FOVs of
Here, we give two properties of sensor selection in our one sensor Si and its sensing probability model. The dashed
binary sensor system. lines segment the space into three small regions. In the gray
Theorem 1: Given a subset of m sensors, U = region, targets can be detected. The dotted lines are the cen-
(S1 , S2 , . . . , Sm ), the information gain (i.e., mutual informa- ter lines of each small regions. The exact 1 or 0 occurs on
tion) is a summation of all CIGs, and it is upper bounded by the dotted lines. Assume that the space is also gridded into
entropy H(X). N uniformly distributed grids. It is known that each grid is
Proof: Based on the chain rule of mutual information, we associated with one binary code. The probability of grid gi
have with the code ci = 1 is P(ci = 1|gi ) = (/( + )), where
and can be the angles or the circular arcs with the origin at
I(X; U) = I(X; S1 ) + I(X; S2 |S1 ) + I(X; S3 |S1 , S2 ) +
the sensor, as shown in Fig. 7. The smaller is, the higher
+ I(X; Sm |S1 , S2 , . . . , Sm1 ) detection probability we can gain. We also denote the detec-
= H(X) HU (X|Cm ) H(X). (13) tion ratio = (/). In this way, the whole tracking space
can be represented by a probability matrix Vi , where

Theorem 2: Let U1 and U2 be two sets with different sensor 11 1nc
orders from the subset U. Then . .. ..
Vi =
.. .
. . (15)
I(X; U1 ) = I(X; U2 ) = I(X; U). (14)
nr 1 nr nc
Proof: By Theorem 1, the mutual information between
targets and sensor sets U1 and U2 can be written as The nr and nc are the number of rows and columns, respec-
tively. For simplicity, let i indicate the detection ratio on
I(X; U1 ) = H(X) H(X|U1 ) grid gi . By updating the probability matrix, the parameters,
I(X; U2 ) = H(X) H(X|U2 ). orientation, and position of one sensor can be calibrated.
Since right-side terms of the two equations are the same,
we can conclude that I(X; U1 ) = I(X; U2 ). B. Sensor Calibration
In our problem, the sensor selection is a dependent process. Assume two grids g1 and g2 are selected from both sides
Based on Theorems 1 and 2, we can see that our CIG process of the middle dotted line, as shown in Fig. 8(a). Let be
is not a Markov chain. That means steps i in the process needs the fixed angle which represents the angle of each small FOV
all the information of previous steps from 1 to i 1 instead region. 1 is the angle between g1 and the middle dotted line,
of just step i 1. and 2 is the angle between g2 and the middle dotted line.
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

6 IEEE TRANSACTIONS ON SYSTEMS, MAN, AND CYBERNETICS: SYSTEMS

Algorithm 1: Self-Calibration
Input: Binary measurements B, calibration paths P,
number of sensors m, and sampling geometry
G(, )
Output: ,
1 i 1;
2 while i < m do
3 Initialization: , ;
4 Wi , Gi PredictedProbFunc(P, gi (i , i ));
Fig. 8. Illustration of calibration with parameters. (a) Orientation. /* initialize */
(b) Positions. 5 Vi BinToProbFunc(Bi , P) ;
/* measurement */
6 Vi Wi Gi ; /* prediction */
Then is equal to 1 + 2 . When introducing the sensing 7 ei
Vi Vi
2 +
Wi
+
Gi
;
probability model and the detection ratio, we can rewrite as 8 while ei not convergent do
  V GT
1 2 9 Wi Wi Vi GiT ;
= + (16) i i
1 + 1 1 + 2 WT V
10 Gi Gi W Ti V Ti ;
i i
where 1 = (1 /( 1 )) and 2 = (2 /( 2 )). Note 11 end
that (1 /(1 + 1 )) is equal to 1 and (2 /(1 + 2 )) is 12 Vi Wi Gi ; /* reconstruction */
equal to 2 . 13 i , i CalibrationFunc(Vi );
Once we have 1 or 2 and the known positions of g1 or g2 , 14 i + +;
we can calculate the rotation angle i of the sensor. However, 15 end
the position of the sensor Si can be any location on the blue
circle. To calibrate the positions i of sensor Si , we select two
more grids g3 and g4 from both sides of the middle dotted line,
as shown in Fig. 8(b). Then, we have the triangle relationship VI. FACTOR G RAPH -BASED D ISTRIBUTED
M ESSAGE PASSING
2
d1S i
+ d2S
2
i
2d1Si d2Si cos(12 ) = d12
2
(17) To verify the efficiency of the proposed sensor selection
2
d3S i
+ d4S
2
i
2d3Si d4Si cos(34 ) = 2
d34 (18) scheme (refer back to Section IV) as well as the accuracy of
the sensor calibration (Section V), we need to apply them to
where dnSi denotes the distance between grid gn and sensor practical multihuman tracking applications and test the track-
Si , n = (1, . . . , 4). d12 denotes the distance between two grids ing effect to see if the proposed schemes can reduce tracking
g1 and g2 ; 12 is the angle that two grids g1 and g2 and errors. Therefore, in this section, we will provide our factor
sensor Si hold, which is represented as g1 Si g2 and can be graph-based message passing model, which is the critical part
calculated from (16). Then, the sensor Si can be located at the in our tracking scheme.
intersection point of blue and red circles. Because we have
known the positions of the four grids, by using (17) and (18), A. Factor Graph
the positions i of the sensor Si can be computed.
Fig. 9(a) shows the directed graph model-based multiple
target tracking with unknown data association. The hidden
C. Self-Calibration variable Vt is a Ot K measurement-to-tracker association
It is clear that if we know the measured probability matrix matrix, and is a multidimensional (Ot ) one-out-of-K random
Vi of the sensor Si , we can calibrate the parameters of the variable. Pd and Pfa are the probabilities of detection and false
sensor, as shown in Section V-B. In real applications, it is alarms, respectively. The joint probability function of X, Z,
time-consuming to cover all the grids when doing calibration and V is given by
process. Thus, we only use a few measurements in sensing p(X, Z, V) = p(Z|X, V)p(V)p(X)
probability matrix of a sensor. In other words, we can calibrate
T 
T
all sensors with a few trajectories. = p(Zt |Xt , Vt ) p(Vt )
We can use a matrix factorization to update the probability t=1 t=1
basis matrix G. The updated weighting matrix and sensing

T1
probability basis matrix can be used as initial matrices in p(Xt+1 |Xt ) p(X1 ) (19)
self-calibration algorithm. They can be obtained by using the t=1
information of predicted initial sampling geometry G(, )  t K j k Vtjk
and calibration paths P. Note that the binary readings can be where p(Zt |Xt , Vt ) = O
j=1 k=1 p(zt |xt ) .
converted into estimated probability measurement matrix V In a distributed PIR sensor system, the whole tracking pro-
based on the calibration paths P. An implementation of the cess can be seen as a message passing process. Therefore, we
self-calibration method is given in Algorithm 1. change the directed graph model into undirected factor graph.
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

LU et al.: PREPROCESSING DESIGN IN PIR SENSOR-BASED HUMAN-TRACKING SYSTEM 7

Fig. 10. Illustration of (a) two layer PIR sensor node (sensor board and iris
Fig. 9. Illustration of (a) directed graph model and (b) undirected factor mote) and a wireless PIR sensor node with two coded masks and (b) PIR
graph model for tracking. sensor system testbed for multiple human tracking.

Fig. 9(b) shows the undirected factor graph of graphical model 2) For the messages flowing into variable nodes Vt and Xt
converted from directed graph model in Fig. 9(a).
Zt ft2 = 1 (31)
Based on (19), the potential functions of joint factors in the 
factor graph are ft2 Xt = ft2 Vt ft2 Zt ft2 f 3 Xt (32)
t
Vt
ft1 = (Vt ) = p(Vt ) (20) 
ft2 Vt = ft2 Xt ft2 Zt ft2 ft1 Vt . (33)
= (Zt , Xt , Vt ) = p(Zt |Xt , Vt )
ft2 (21)
Vt
p(X1 ), t = 1
ft = (Xt ) =
3
(22) 3) For the message flowing from Xt to Xt+1
1, t>1
(Xt+1 , Xt ) = p(Xt+1 |Xt ). (23) Xt Xt+1 = Xt1 Xt ft2 Xt f 3 Xt . (34)
t

The belief propagation message passing updates involve two


B. Distributed Message Passing operations: 1) multiplication and 2) summation. The multipli-
The target dynamic model and sensor observation model cation step is used to aggregate information from neighbor
can be written as nodes, as shown in (7). The summation step is used to
marginalize a certain node on the factor graph, as shown in (8).
Xt+1 = AXt + t
Zt = HXt + t (24) VII. P ERFORMANCE A NALYSIS
where and are zero-mean noises with = Q and E{T } A. Experiment Setup
E{ T } = R. For simplicity, we omit the time index t unless To verify the proposed distributed binary sensor system with
it is necessary sensor selection and calibration, we have developed the PIR
f 1 = (V) = p(V) sensor system testbed for multiple human tracking, as shown
in Fig. 10(b). Our PIR sensor-based binary sensing system
= PD (1 PD )K Pfa (1 Pfa ) (25) is a nonintrusive, scalable, and low-cost system. Fig. 10(a)
f = (Z, X, V) = p(Z|X, V)
2
shows the two-layer PIR sensor node with coded masks. The

O sensor board contains two PIR sensors, a signal conditioning
= N (z(i); Hx(V(i)), R) (26) circuit, and a programmable system on chip. The Iris mote,
i
with RF transceiver, is connected to sensor board through a
p(X1 ) = N (X1 ; AX0 , Q0 ), t = 1 51-pin connector. The sampling rate is 9 Hz. The PIR sensors
f 3 = (X) = are deployed on the walls to measure human presence. Results
1, t>1
are displayed on two big screens for visualization purpose.
(27) Fig. 11 illustrates the tracking experiment setup.
(Xt+1 , Xt ) = p(Xt+1 |Xt ) = N (Xt+1 ; AXt , Qt ) (28) Experiments are conducted in a 6 m 6 m observa-
tion space where 144 plastic floor tiles are deployed. The
where is the number of detections; is the number of false
size of each mat is about 0.5 m 0.5 m. Sixteen PIR sensor
alarms in ; O is the number of observations; x(V(i)) denotes
nodes (32 sensors) are deployed on four walls with pegboards.
the target state associated with observation i; and X0 and Q0
These sensor nodes are easy to move and are scattered so
are initial values.
that each small region of observation space can be detected.
We use belief propagation to achieve distributed message
The sensor nodes are also placed to face the center of the
passing. Here, we provide the messages in the factor graph as
observation space. Once they are on the walls, the predicted
shown in Fig. 9(b).
orientations and positions of the sensors are known.
1) For the messages flowing out of variable
nodes Vt and Xt
B. Sensor Selection
Vt ft2 = ft1 Vt = (Vt ) (29) It is important to select the sensors not only to reduce the
Xt ft2 = Xt+1 Xt Xt1 Xt f 3 Xt . (30) amount of data but also to improve the fusion accuracy by
t
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

8 IEEE TRANSACTIONS ON SYSTEMS, MAN, AND CYBERNETICS: SYSTEMS

Fig. 11. Illustration of tracking (a) one target case, (b) two targets case, and
(c) three targets case.

Fig. 13. Illustration of tracking errors with minimum information gain,


maximum information gain, and best sensor selection processes at 4, 6, and
12 sensor nodes.

bound I(X; U) I(X; U ), where U can be an arbitrary subset


that contains m sensor nodes. It is obviously that if we can
do explicit search in each step of sensor selection, it is guar-
anteed that we can select the optimal set of sensor nodes.
However, it is time-consuming and computational complex.
In our approach, our selection relies on previous selections,
which is a conditional process. We can see that, for most of
time the maximum and minimum information gain-based sen-
sor selection processes hold the lower and upper bounds of
sensor selection process, as shown in Fig. 12(a). Thus, our
information-based sensor selection approach is very close to
the optimal solution.
Fig. 12(b) shows the mutual information with respect to
different sensor node numbers for one target. The red curve
here is the sensor selection process with the maximum mutual
Fig. 12. (a) Information gain process for one target. (b) Mutual information information at each sensor selection step. The blue curve here
process for one target. (c) Information gain process for two targets. (d) Mutual
information process for two targets. is the sensor selection process with the minimum mutual infor-
mation at each sensor selection step. The solid line is H(X). It
is the ceiling bound of mutual information I(X; U), where U
selecting the most reliable sensors for a certain region at a includes all the sensor nodes. It shows the similar results
certain time. Sensor selection is proposed here to solve such in Fig. 12(a). The more sensor nodes we select, the higher
problem. mutual information between target(s) and sensor nodes we will
Fig. 12(a) shows the information gain with respect to dif- get. Fig. 12(c) and (d) shows the information gain and mutual
ferent numbers of sensor nodes for one target. The red curve information with respect to different numbers of sensor nodes
is the sensor selection process with the maximum informa- for two targets.
tion gain at each sensor selection step. The blue curve is the Fig. 13 illustrates the tracking errors with three different
sensor selection process with the minimum information gain sensor selection processes. They are minimum information
at each sensor selection step. The green curve is the sensor gain process, maximum information gain process and best
selection process with other information gains at each sensor selection process. The best selection uses the lowest infor-
selection step. Fifty green curves have been generated in our mation gain for a given number of sensor nodes. One target
experiment. It shows that with the maximum information gain walks along a circle path for this experiment. No calibration
in each sensor selection step, the information gain between is provided here. For comparison, we use 4, 6, and 12 sensor
the previous sensor set and the current one is reduced faster. nodes with the best selection scheme. It can be seen that max-
On the other hand, with the minimum information gain in imum information gain process in sensor selection provides a
each sensor selection step, the information gain between the very close performance to best selection. In fact, in Fig. 12,
previous sensor set and the current one is reduced slower. all the  points are very close to red curve. This result indi-
It also shows that, for numbers 4, 9, 12, 13, 14, and 15, some cates that maximum information gain sensor selection process
of the information gains from other sensor selection processes can give close to the best performance of tracking. It is also
are larger than the one shown in blue curve, marked out as  obvious to see that the more sensors we selected, the better
in Fig. 12(a). Similar results also show that some of the infor- tracking performance we could get.
mation gains at 4, 6, and 12 are below the red curve, marked
as . These results indicate an boundary assumption. For a C. Subregion Sensor Selection
given number m of sensor nodes, let U = (S1 , S2 , . . . , Sm ) Subregion sensor selection (see Section IV-C) uses the
be the subset of sensor nodes selected by information gain- locations of targets to narrow down the regions that are
based process, there exists a subset U that holds the used to calculate information gain or mutual information.
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

LU et al.: PREPROCESSING DESIGN IN PIR SENSOR-BASED HUMAN-TRACKING SYSTEM 9

Fig. 15. Comparisons of tracking errors and mutual information with respect
to whole-space and subregion sensor selection processes.

Fig. 14. Illustration of Voronoi diagram of (a) subregion sensor selection


for one target, (b) subregion sensor selection for two targets, (c) whole-space
sensor selection I for one target, and (d) whole-space sensor selection II for
one target.

Fig. 16. Illustration of mobile calibration.


Fig. 14(a) and (b) shows the Voronoi space segment plots using
subregion sensor selection method with respect to one target
and two targets cases. It can be seen that, with the target loca-
tion information, we can provide better space segment, which
means smaller segment units. This method focuses on sub-
regions of tracking space to maximize the information gain.
On the other hand, Fig. 14(c) and (d) gives the Voronoi space
segments with whole-space sensor selection method. Fig. 14(c)
provides the same mutual information as in Fig. 14(a). In this
case, the number of segment units is close to the number of
segment units in subregion in Fig. 14(a) while the sizes are
larger. Thus, the tracking error is larger. However, the number Fig. 17. Illustration of mobile calibration draws and sensing probability
of sensor nodes is smaller. Fig. 14(d) illustrates the similar matrices. (a) and (e) Diagonal path case. (b) and (f) Full coverage case.
(c) and (g) Three-path case. (d) and (h) Four-path case.
tracking error as in Fig. 14(a). In this case, the number of
sensor nodes used to achieve such tracking performance is
larger although the mutual information is larger. Fig. 14(b)
give good tracking resolution. From the above result analysis,
shows the Voronoi space segment plot with two targets using
we can see that the advantages of subregion sensor selection
subregion sensor selection method. Similar conclusions to one
method are: 1) with the same mutual information, it gives the
target case (discussed above) can be drawn here, too.
better tracking performance without using many sensor nodes
Fig. 15 gives the comparisons of tracking errors and mutual
and 2) with the same tracking performance, it uses the less
information with respect to whole-space and subregion sen-
number of sensor nodes.
sor selections. One target walks along a circle path. In total,
412 sensor nodes are used to perform tracking. It can be
seen that the mutual information of subregion sensor selection D. Mobile Calibration
process is lower than the mutual information of whole-space. An Android mobile calibration application is developed to
At number of sensor nodes 4 and 8, the mutual information customize the calibration paths and control tracking process
for subregion and whole-space, 4.0219 and 3.8611, is close remotely. As shown in Fig. 16, a subject can preset one path
to each other. However, the tracking error of subregion is in the mobile device by drawing line(s), then click calibration
0.2677 m less than the tracking error of whole-space. Also, the start button and walk along the preset path to do calibration
tracking errors are reduced and close to each other when the for sensors. If more paths are needed for better calibration, one
number of sensor nodes are increased. For the tracking error can draw one path and do calibration one by one. It can be
at around 0.56 m, subregion method selects six sensor nodes done simply with five steps: 1) stop tracking; 2) draw a cali-
while whole-space method selects nine sensor nodes although bration path; 3) start calibration and walk along the calibration
it gives higher mutual information. We also notice that, for path; 4) stop calibration; and 5) start tracking (skip step 1) if
one target case, the number of sensor nodes larger than eight new tracking experiment is conducted.
has small tracking accuracy improvement. So, it is reason- Fig. 17(a)(d) shows the four different customized calibra-
able to select eight sensor nodes in sensor selection process to tion paths drawn for mobile applications. The gridded areas
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

10 IEEE TRANSACTIONS ON SYSTEMS, MAN, AND CYBERNETICS: SYSTEMS

Fig. 19. Illustration of the true trajectories. (a) Diagonal path. (b) Circle
path. (c) Random path.
Fig. 18. Illustration of calibration errors of the sensor parameters at 6 oclock
using full coverage calibration.

are the tracking space paved by plastic interlocking floor


tiles, as shown in Fig. 16 (right). There are 12 12 grids.
Fig. 17(a) and (b) is one-path calibration draws. However,
Fig. 17(b) covers all the tracking space in order to do cali-
bration. Fig. 17(c) and (d) is the multipath calibration draws.
Compared with Fig. 17(a), (c), and (d) gives more coverage.
Compared with Fig. 17(b), they reduce the calibration time by
only walking along several short paths.
After setting calibration paths, the predicted sensing proba-
bility matrix will be generated with values only related to the
calibration paths. Fig. 17(e)(h) shows the sensing probabil- Fig. 20. Illustration of the random path. (a) True trajectories in X locations.
ity matrices for the sensor located on the 6 oclock position (b) True trajectories in Y locations. (c) Tracking errors.
with respect to four different calibration paths. During calibra-
tion, the binary readings of each sensor are converted into the
measured sensing probability matrix. By using the calibration calibration gives better tracking performance than diagonal and
process discussed in Section V, we are able to estimate the three-path calibrations. It is 0.0686 m larger than the full cov-
sensing probability matrix, and then calibrate the positions and erage calibration. For the random trajectory, the two multipath
orientations of the sensor. The calibrated parameter errors of calibrations give close performances, 0.6193 and 0.5598 m.
the sensor at 6 oclock with full coverage calibration are given However, tracking error of the full coverage calibration can
in Fig. 18. The green star is the initial prediction of sensor reach 0.3347 m. Furthermore, tracking errors at different times
parameters. The red star is the calibrated sensor parameters. are given in Fig. 20 for a random trajectory which can be
Note that the original assumption of this sensors parameters is divided into three short trajectories. We use the diagonal path
( = 90 , x = 3 m, y = 0 m). The calibrated parameters, for calibration. The average tracking error is 0.5756 m. It is
( = 4 , x = 0.13 m, y = 0.18 m), mean that the sen- shown that during the diagonal trajectory, the tracking error
sor rotates 4 clockwise to 86 , shifts left 0.13 m to 2.87 m is small while the other two curved trajectories have higher
and moves forward 0.180.18 m. tracking errors. This result further indicates that if we know
the target trajectory, we can customize the calibration path to
E. One Target Tracking With Calibration give a better tracking performance unless the full coverage
is used. In fact, in Figs. 13 and 15, without calibration, the
In order to discuss the improvements in tracking perfor-
best tracking performance is around 0.55 m for circle trajec-
mance of the four customized calibration paths given above,
tory. After using circle path calibration, the tracking error can
the three trajectories (diagonal, circle, and random paths) are
be reduced to 0.32 m. Our proposed calibration method can
tested for one target case. Here, the number of sensor nodes
greatly improve tracking performance in our binary system.
to be selected is 8 and the maximum information gain process
with subregion method is used. The three true trajectories are
shown in Fig. 19. F. Multiple Target Tracking
Table I provides the results of tracking errors with dif- In this section, we consider the tracking problem of multi-
ferent trajectories and calibrations. It can be perceived that ple subjects who interact with each other over a period of
the performance of tracking after calibration is dependent time. The number of targets is known a priori. Thus, we
on the trajectories of the moving subjects. For all three tra- do not consider track initialization and termination here. We
jectories, the full coverage calibration has the best tracking consider the three-target scenario case. The results are shown
performances. For the diagonal trajectory, the diagonal cali- in Fig. 21. A full coverage calibration is used with all sensor
bration has smaller tracking error than multipath calibrations. nodes selected. It can be seen that although we have used 16
The tracking error, 0.3573 m is close to the lowest track- sensor nodes and full coverage calibration, the tracking reso-
ing error 0.3024 m. For the circle trajectory, the four-path lution is not very high but acceptable. In the given time slots,
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

LU et al.: PREPROCESSING DESIGN IN PIR SENSOR-BASED HUMAN-TRACKING SYSTEM 11

TABLE I
T RACKING E RRORS A MONG D IFFERENT T RAJECTORIES AND D IFFERENT C ALIBRATION PATHS

Fig. 22. Illustration of tracking resolutions with sampling rates. (a) 11 Hz.
(b) 17 Hz.

and the tracking resolution will be better. Fig. 22 illustrates


the tracking resolutions with different sampling rates for three-
target scenario. At sampling rate 11 Hz, the average tracking
error is 0.6861 m. When the sampling rate increases to 17 Hz,
Fig. 21. Three-target tracking scenario with 16 sensors and full coverage the average tracking error is reduced to 0.4706 m. However,
calibration. (a) X locations of three paths with time. (b) Y locations of three
paths with time. (c) True trajectories. (d) Estimated trajectories.
the computational complexity of the message-passing algo-
rithm would be exponentially increased. Fig. 23 shows the
relationship between running time and sampling rate for three-
circled in Fig. 21(c) and (d), the three targets are very close target scenario given above. As we can see, the running time
to each other. The tracking errors during these time periods of our message passing algorithm is getting longer when the
are higher. sampling rate is increasing. Running time > 1 s will cause
The message passing algorithm relies on the number of mes- delay in tracking. Longer delay (> 2 s) is not acceptable.
sages passed in different times. If we increase the sampling Thus, we set the sampling rate to be less than or equal
rate of sensing, more messages will be passed in the network to 17 Hz.
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

12 IEEE TRANSACTIONS ON SYSTEMS, MAN, AND CYBERNETICS: SYSTEMS

[3] J. He, Y. Geng, and K. Pahlavan, Toward accurate human tracking:


Modeling time-of-arrival for wireless wearable sensors in multipath envi-
ronment, IEEE Sensors J., vol. 14, no. 11, pp. 39964006, Nov. 2014.
[4] K.-C. Lan and W.-Y. Shih, Using smart-phones and floor plans for
indoor location tracking, IEEE Trans. Human-Mach. Syst., vol. 44,
no. 2, pp. 211221, Apr. 2014.
[5] Q. Hao, F. Hu, and J. Lu, Distributed multiple human tracking with
wireless binary pyroelectric infrared (PIR) sensor networks, in Proc.
IEEE Sensors, Kona, HI, USA, Nov. 2010, pp. 946950.
[6] J. Lu, J. Gong, Q. Hao, and F. Hu, Space encoding based compressive
multiple human tracking with distributed binary pyroelectric infrared
sensor networks, in Proc. IEEE Conf. Multisensor Fusion Integr. Intell.
Syst., Hamburg, Germany, Sep. 2012, pp. 180185.
[7] Y. Bar-Shalom and T. E. Fortmann, Tracking and Data Association.
Boston, MA, USA: Academic, 1987.
Fig. 23. Relationship between running time and sampling rate for three-target [8] R. Ma and Q. Hao, Buffons needle model based walker recogni-
scenario by using message passing algorithm. tion with distributed binary sensor networks, in Proc. IEEE Conf.
Multisensor Fusion Integr. Intell. Syst., Hamburg, Germany, Sep. 2012,
pp. 120125.
[9] J. Lu, J. Gong, Q. Hao, and F. Hu, Multi-agent based wireless
VIII. C ONCLUSION pyroelectric infrared sensor networks for multihuman tracking and
self-calibration, in Proc. IEEE Conf. Sensors, Baltimore, MD, USA,
In this paper, we have presented a multiple human track- Nov. 2013, pp. 14.
ing system with sensor selection and calibration based on [10] Q. Sun, F. Hu, and Q. Hao, Mobile target scenario recognition via low-
distributed binary PIR sensors. The information-based sensor cost pyroelectric sensing system: Toward a context-enhanced accurate
identification, IEEE Trans. Syst., Man, Cybern., Syst., vol. 44, no. 3,
selection process has been developed to select the best subset pp. 375384, Mar. 2013.
of sensor nodes. This procedure provided suboptimal upper [11] Q. Sun, Z. Wu, J. Lu, F. Hu, and K. Bao, Low-cost pyroelectric sen-
and lower bounds for mutual information between targets and sor networks for Bayesian crowded scene analysis, in Proc. IEEE Int.
Conf. Mobile Ad-hoc Sensor Netw. (MSN), Maui, HI, USA, Dec. 2014,
sensor nodes. Target locations were also used in sensor selec- pp. 8895.
tion to perform the subregion sensor selection. The sensing [12] S. B. Crary and Y. Jeong, Bayesian optimal design of experiments
probability model was introduced in calibration. A matrix for sensor calibration, in Proc. IEEE Int. Conf. Solid-State Sensors
Actuators Eurosensors IX. Transducers, vol. 2. Stockholm, Sweden,
factorization method was used to update the sensing proba- Jun. 1995, pp. 4851.
bility model. And the calculation of sensor parameters from [13] C. Giraud and B. Jouvencel, Sensor selection: A geometrical approach,
the updated probability model was also provided. Distributed in Proc. IEEE Int. Conf. Intell. Robots Syst., vol. 2. Pittsburgh, PA, USA,
Aug. 1995, pp. 555560.
message passing framework was discussed with a factor graph [14] R. Garnett, M. A. Osborne, and S. J. Roberts, Bayesian optimization
model for target tracking. The experiment results demon- for sensor set selection, in Proc. 9th ACM/IEEE Int. Conf. Inf. Process.
strated the good tracking performance after sensor selection Sensor Netw., Stockholm, Sweden, 2010, pp. 209219.
[15] K. Ni and G. Pottie, Bayesian selection of nonfaulty sensors, in Proc.
and calibration. IEEE Int. Symp. Inf. Theory, Nice, France, Jun. 2007, pp. 616620.
The advantages of binary PIR sensor system in tracking [16] F. Zhao, J. Shin, and J. Reich, Information-driven dynamic sensor
include flexibility, scalability, robustness, and inexpensiveness. collaboration, IEEE Signal Process. Mag., vol. 19, no. 2, pp. 6172,
Mar. 2002.
Such sensing system can be used in smart indoor tracking [17] H. Wang, K. Yao, and D. Estrin, Information-theoretic approaches for
applications. Our proposed system has demonstrated satisfac- sensor selection and placement in sensor networks for target localiza-
tory tracking results with sensor selection and calibration. The tion and tracking, IEEE J. Commun. Netw., vol. 7, no. 4, pp. 438449,
Dec. 2005.
resolution of tracking can reach 0.35 m. In the future, we will [18] X. Shen and P. K. Varshney, Sensor selection based on generalized
further improve the accuracy from the following two aspects. information gain for target tracking in large sensor networks, IEEE
1) Improve the design of the binary compressive measure- Trans. Signal Process., vol. 62, no. 2, pp. 363375, Jan. 2014.
[19] D. Bajovic, B. Sinopoli, and J. Xavier, Sensor selection for event detec-
ment matrix. The encoded space can be redesigned to tion in wireless sensor networks, IEEE Trans. Signal Process., vol. 59,
reduce the number of repeated binary codes. We will no. 10, pp. 49384953, Oct. 2011.
also design the binary codes (e.g., LDPC) to maximize [20] Y. Zhang and Q. Ji, Efficient sensor selection for active information
fusion, IEEE Trans. Syst., Man, Cybern. B, Cybern., vol. 40, no. 3,
coding efficiency and minimize the number of sensors. pp. 719728, Jun. 2010.
2) Develop a better approach for measurement decoding. [21] J. Feng, S. Megerian, and M. Potkonjak, Model-based calibration for
In distributed multiple target tracking scenarios, sen- sensor networks, in Proc. IEEE Sensors, vol. 2. Toronto, ON, Canada,
2003, pp. 737742.
sor readings need to be decoded into binary codes. [22] J. Feng, G. Qu, and M. Potkonjak, Actuator-based infield sensor
A decoding optimization approach is needed to have low calibration, IEEE Sensors J., vol. 6, no. 6, pp. 15711579, Dec. 2006.
decoding error. [23] F. Koushanfar, M. Potkonjak, and A. Sangiovanni-Vincentelli, Error
models for light sensors by statistical analysis of raw sensor measure-
ments, in Proc. IEEE Sensors, Vienna, Austria, 2004, pp. 14721475.
[24] K. Whitehouse and D. Culler, Calibration as parameter estimation in
R EFERENCES sensor networks, in Proc. ACM Int. Workshop Wireless Sensor Netw.
Appl., Atlanta, GA, USA, 2002, pp. 5967.
[1] H. T. Cheng and W. Zhuang, Bluetooth-enabled in-home patient mon- [25] L. Balzano and R. Nowak, Blind calibration of sensor networks,
itoring system: Early detection of Alzheimers disease, IEEE Trans. in Proc. 6th Int. Conf. Inf. Process. Sensor Netw., Cambridge, MA,
Wireless Commun., vol. 17, no. 1, pp. 7479, Feb. 2010. USA, 2007, pp. 7988.
[2] A. Colombo, D. Fontanelli, D. Macii, and L. Palopoli, Flexible indoor [26] S. Bolognani, S. Del Favero, L. Schenato, and D. Varagnolo,
localization and tracking based on a wearable platform and sensor Consensus-based distributed sensor calibration and least-square param-
data fusion, IEEE Trans. Instrum. Meas., vol. 63, no. 4, pp. 864876, eter identification in WSNS, Int. J. Robust Nonlin. Control, vol. 20,
Apr. 2014. no. 2, pp. 176193, 2010.
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

LU et al.: PREPROCESSING DESIGN IN PIR SENSOR-BASED HUMAN-TRACKING SYSTEM 13

[27] N. Wiberg, Codes and decoding on general graphs, Ph.D. dissertation, Fei Hu received the Ph.D. degree in signal pro-
Dept. Electr. Eng., Linkping Univ., Linkping, Sweden, 1996. cessing from Tongji University, Shanghai, China,
[28] N. Wiberg, H.-A. Loeliger, and R. Kotter, Codes and iterative decoding in 1999, and the Ph.D. degree in electrical and
on general graphs, Eur. Trans. Telecommun., vol. 6, no. 5, pp. 513525, computer engineering from Clarkson University,
1995. Potsdam, NY, USA, in 2002.
[29] R. M. Tanner, A recursive approach to low complexity codes, IEEE He is currently a Professor with the Department
Trans. Inf. Theory, vol. 27, no. 5, pp. 533547, Sep. 1981. of Electrical and Computer Engineering, University
[30] R. G. Gallager, Low-density parity-check codes, IEEE IRE Trans. Inf. of Alabama, Tuscaloosa, AL, USA. His research
Theory, vol. 8, no. 1, pp. 2128, Jan. 1962. interests include security: this is about how to over-
[31] V. Indelman, S. Williams, M. Kaess, and F. Dellaert, Factor graph based come different cyber attacks in a complex wireless
incremental smoothing in inertial navigation systems, in Proc. IEEE Int. or wired network. Recently he focuses on cyber-
Conf. Inf. Fusion (FUSION), Singapore, 2012, pp. 21542161. physical system security and medical security issues, signals: this mainly
[32] H. Wang, J. Sun, S. T. Lu, and S. Wei, Factor graph aided multiple refers to intelligent signal processing, that is, using machine learning algo-
hypothesis tracking, Sci. China Inf. Sci., vol. 56, no. 10, pp. 16, 2013. rithms to process sensing signals in a smart way in order to extract patterns,
[33] V. P. Panakkal and R. Velmurugan, Effective data association scheme and sensors: this includes micro-sensor design and wireless sensor networking
for tracking closely moving targets using factor graphs, in Proc. IEEE issues.
Nat. Conf. Commun. (NCC), Bengaluru, India, Jan. 2011, pp. 15.

Jiang Lu (M15) received the B.Sc. degree from


Shanghai Maritime University, Shanghai, China, in
2008, the M.Sc. degree from the University of
Florida, Gainesville, FL, USA, in 2007, and the
Ph.D. degree from the University of Alabama,
Tuscaloosa, AL, USA, in 2015, all in electrical and
computer engineering.
He is currently an Assistant Professor with the
Department of Computer Engineering, University
of Houston at Clear Lake, Houston, TX, USA.
His research interests include intelligent sensor sys-
tems, biomedical devices, tele-healthcare, wireless communication, distributed
information processing, machine learning, and embedded system designs.

Qi Hao received the B.E. and M.E. degrees from


Shanghai Jiao Tong University, Shanghai, China, in
1994 and 1997, respectively, and the Ph.D. degree
from Duke University, Durham, NC, USA, in 2006,
Ting Zhang received the B.E. degree in com- all in electrical engineering.
puter engineering from Tianjing University, Tianjing, His postdoctoral training with the Center for
China, in 2008, and the M.S. degree in computer sci- Visualization and Virtual Environment, University
ence from the University of Texas at Brownsville, of Kentucky, Lexington, KY, USA, was focused on
Brownsville, TX, USA, in 2011. She is currently 3-D computer vision for human tracking and identi-
pursuing the Ph.D. degree in electrical and com- fication. From 2007 to 2014, he was an Assistant
puter engineering with the University of Alabama, Professor with the Department of Electrical and
Tuscaloosa, AL, USA. Computer Engineering, University of Alabama, Tuscaloosa, AL, USA. He
Her research interests include data mining is currently an Associate Professor with the South University of Science
and classification algorithms, computer vision and and Technology of China, Shenzhen, China. His research interests include
computer graphics, tele-healthcare, and mobile smart sensors, intelligent wireless sensor networks, and distributed information
computing. processing.

You might also like