On the Effect of Sensing-holes in PIR-based

Occupancy Detection Systems

Abdelraouf Ouadjaout, Noureddine Lasla, Djamel Djenouri and Cherif Zizoua

CERIST Research Center, Ben-Aknoun, Algiers, Algeria

Keywords:

Wireless Sensor Networks, Building Energy Management Systems, Occupancy Detection, PIR Sensors.

Abstract:

Sensing-holes in PIR-based motion detection systems are considered, and their impact on occupancy moni-

toring applications is investigated. To our knowledge, none of prior works on PIR-based systems consider the

presence of these holes, which represents the major cause for low precision of such systems in environments

featured with very low mobility of occupants, such as working ofﬁces. We consider optimal placement of

PIRs that ensures maximum coverage in presence of holes. The problem is formulated as a mixed integer lin-

ear programming optimization problem (MILP). Based on this formulation, an experimental study on a typical

working ofﬁce has been carried out. The empirical results quantify the effects of the holes on the detection

accuracy and demonstrate the enhancement provided by the optimal deployment of the solution.

1 INTRODUCTION

Detecting user occupancy in buildings is a fundamen-

tal step for reducing wastage of energy and improving

users’ comfort. In fact, in many cases, buildings have

a set of predeﬁned actuation schedules for managing

electrical appliances, such as HVAC and lights. These

schedules have a coarse-grained time dependability

that is generally related to static issues such seasons,

days of the week, etc.. However, by dynamically de-

tecting vacant places, more optimized context-aware

schedules can be implemented that can help shorten-

ing the actuation durations everyday without compro-

mising the users convenience.

Many of the proposed solutions for tracking the

presence of persons in buildings are based on the

use of passive infrared (PIR) sensors (Delaney et al.,

2009; Agarwal et al., 2010; Marchiori and Han, 2010;

Lu et al., 2010; Beltran et al., 2013; Kazmi et al.,

2014). These sensors are made from inexpensive py-

roelectric materials that react to the change of infrared

emissions in the environment, which helps in captur-

ing the presence of humans in a speciﬁc space. The

low cost and low energy consumption of such sen-

sors enable their large use in battery-operated sys-

tems. Further, they do not affect the privacy of people

and do not require the presence of an existing infras-

tructure. This makes them an appropriate candidate

for monitoring private spaces such as ofﬁces, meeting

rooms, etc., where other technologies, such as cam-

eras, cannot be used.

However, a major drawback of PIR sensors is

their false negatives (non-detection) in some situa-

tions. The ﬁrst reason behind this shortcoming is

that these sensors are only capable of detecting mo-

tion, and not static bodies. Whilst this does not repre-

sent any problem in some typical places of a building,

such as corridors and near the doors where people are

generally moving, it prevents the accurate monitoring

in places such as ofﬁces where workers tend to stay

immobile for relatively long periods. To tackle this

problem, some solutions have been proposed in the

literature that complement the PIRs with information

provided by additional sensors. For example, the oc-

cupancy detection system proposed by Agarwal et al.

(Agarwal et al., 2010) is enhanced with a magnetic

reed switch sensor to track the open/close events of

an ofﬁce door and match them with the output of the

PIRs. ThermoSense (Beltran et al., 2013) employs, in

addition to a PIR sensor, a thermal sensor array that

can measure temperatures of a 2.5 m ×2.5 m area dis-

cretized as a 8 × 8 grid. Alternatively, some other so-

lutions use other sensing techniques, such as (Nguyen

and Aiello, 2013).

The second problem is that the sensing area of

a typical PIR module is not a contiguous volume,

but it is featured with the presence of several spaces

where changes of infrared emissions are not captured

by the sensor. We refer to these uncovered spaces

by the term of sensing-holes. The dimensions of

Ouadjaout, A., Lasla, N., Djenouri, D. and Zizoua, C.

On the Effect of Sensing-holes in PIR-based Occupancy Detection Systems.

DOI: 10.5220/0005805901750180

In Proceedings of the 5th International Confererence on Sensor Networks (SENSORNETS 2016), pages 175-180

ISBN: 978-989-758-169-4

175

— — —

0.6 m 1.25 m 2.5 m

Sensitivity

Sensing-hole

size

Hand

motion

Arm

motion

Body

motion

(a) (b)

Figure 1: (a) The relation between the size of sensing-holes and motion sensitivity. (b) 2D view of a PIR Field-of-View. Solid

rectangles represent detection zones.

these sensing-holes become larger as we move far

from the sensor, e.g., it reaches the scale of a human

body movements at a distance of 2 m to 3 m, typical

hight of ceiling at ofﬁces, where PIRs are usually in-

stalled. Consequently, a PIR cannot detect a person

within the sensing-hole even when he performs small

movements (e.g., in the ofﬁce scenario, moving his

arms, rotating the chair when sitting, etc.). While it

seems infeasible to detect static body only with PIRs

(the ﬁrst problem), we think it is possible to tackle

the second one by investigating the sensing-holes and

their impact, and using optimal deployment of PIRs

to eliminate/minimize such holes. This represents the

subject of this work. To the best of our knowledge,

this problem has not been considered previously in

the literature.

The remaining of the paper is organized as fol-

lows, Sec. 2 presents the problem formulation where

the problem of sensing-holes is introduced and PIR

deployment for optimal coverage is modeled as a

mixed integer linear program (MILP). Solution of this

MILP is used in the experimentations, which are pre-

sented in Sec. 3. Finally, Sec. 4 draws the conclusions

and sketches the perspectives.

2 PROBLEM FORMULATION

PIRs use pyroelectric transducers that convert in-

frared radiations into electrical signals. To increase

the PIR sensitivity, a Fresnel lens is used which con-

centrates infrared radiations onto the detector. This

results in a ﬁeld-of-view (FoV) that is more like a dis-

crete set of beams or cones, including many sensing-

holes. To be detected, the movements of the person

should take place within the FoV. Fig. 1(a) illustrates

the different types of motion made by a human and the

corresponding maximum sensing-hole size for which

the motion can be detected by a PIR (California En-

ergy Commision, 1993). The sensing-holes should

not be more than 0.6 m to ensure an efﬁcient detection

by the PIR of a sitting person’s movements caused by

hand motions.

The size and distribution of the holes impact the

granularity of the PIR detections. Fig. 1(b) illus-

trates the projection of the actual FoV of a Pana-

sonic EKMB PIR sensor on a two dimensional plane

(Panasonic, 2012). The PIR is placed at the ceiling

of an ofﬁce and the projection is performed on the

plane parallel to the ground and elevated at a typical

height of desks, where most of persons’ low move-

ment activities take place (e.g. arm and hand move-

ment when sitting). The ﬁgure shows the presence of

several sensing-holes that represent more than 87%

of the total monitored ofﬁce area, and their sizes vary

from one region to another within the PIR’s FoV and

can exceed 1 m in some places. These large sensing-

holes can affect PIR-based occupancy detection sys-

tems and cause incorrect decisions to be taken, such

as turning off a light or HVAC in the presence of a

person, which limit the credibility of the system.

To alleviate the negative impact of the sensing-

holes, we deﬁne the Maximal PIR Coverage (MPC)

problem that ﬁnds the optimal positions of the PIRs

for maximum coverage in the area of interest. To sim-

plify the problem, we do not consider the 3D coverage

but the study herein is limited to the projection of the

covered area on a two dimensional plane as explained

before. Even with such simpliﬁcation, the computa-

tion of the union of the detection zones for a given

set of PIRs is difﬁcult to formulate mathematically.

Therefore, we discretize the monitored area and con-

sider it as a set of points, where a point will be con-

sidered covered iff it is within the coverage zone of

SENSORNETS 2016 - 5th International Conference on Sensor Networks

176

(a) (b)

(c)

Figure 2: Three deployments scenarios considered during the experimentation. Circle points represent the discrete grid of

the deployment area and diamond points represent the PIRs. The gray zone delimits the place of the ofﬁce’s desk, which has

been given a greater weight in Φ. (a) Optimal deployment with one PIR. (b) Optimal deployment with three PIRs ensuring

full coverage of the desk’s area. (c) Hole-unaware deployment where the desk area is completely uncovered.

at least one PIR. Formally, let D denotes the two di-

mensional space to be monitored by a set, S, of PIRs.

For the sake of simplicity, we assume that D has a

rectangular shape with width W and length L. With-

out loss of generality, we discretize D by dividing its

sides with a step l, which results in a grid of points

D = {(i.l +

, j.l +

) | 0 ≤ i ≤ b

c ∧ 0 ≤ j ≤ b

c}.

In general, the density of persons in the space

is not homogeneously distributed and obstacles (e.g.,

bookshelf, desks, table, etc.) may also be present.

Therefore, we introduce a weighting matrix Φ

(x,y)∈

that enriches the geometric deployment space with a

semantic dimension indicating the places where peo-

ple are more likely to be detected. This is by given a

high weight to areas where people are likely to stay

and thus exhibit low movement activity once there

(e.g., area of a desk chair, meeting table chairs), zero

weight at obstacles (e.g., bookshelf, table, etc.), and

regular weight elsewhere where people are likely to

move. We assume that PIRs are placed on the ceil

of the deployment area without any rotation. Con-

sequently, the detection zones of any PIR s ∈ S will

have a rectangular shape and can be modeled by a set

⊆ R

, where a tuple (x

, x

, y

) ∈ Z

denotes the

boundaries of a single detection zone, z

, on the X and

Y , when s is placed at the origin (0, 0). The conse-

quence of a change in the PIR coordinate from the

origin, say to the position (X

, Y

), is a simple transla-

tion of the zone, z

, on the abscissa and ordinate axes,

by, X

, and Y

, respectively.

The MPC problem can then be formalized as a

mixed integer linear problem (MILP) as follows. We

deﬁne the decision variables X

and Y

for denoting

the coordinates of a PIR s ∈ S, and the binary de-

cision variables C

(x,y)

that indicate wether the point

(x, y) ∈

D is covered by at least one PIR or not. The

objective function can therefore be deﬁned as:

max

∑

(x,y)∈

(x,y)

× Φ

(x,y)

, (1)

with the following two constraints:

∀(x, y) ∈

D :

(x,y)

= 0∨

∃s ∈ S, ∃(x

, x

, y

) ∈ Z

+ x

≤ x ≤ X

+ x

)∧

+ y

≤ x ≤ Y

+ y

)

l (2)

and,

∀s ∈ S :

(0 ≤ X

≤ W ) ∧ (0 ≤ Y

≤ L).

(3)

On the Effect of Sensing-holes in PIR-based Occupancy Detection Systems

177

The ﬁrst constraint formalizes that C

(x,y)

= 1 in the

solely case of the existence of at least one detection

zone of a sensor s that covers the point (x, y), while the

second restricts the coordinates of the sensors within

the deployment area. To eliminate the operators ∃ and

∨ in (2), and transform the MILP into a standard form

that can be handled by solvers, the big-M method can

be used.

3 EXPERIMENTS

We have deployed an experimental PIR-based occu-

pancy detection system to monitor an ofﬁce and quan-

tify the impact of the sensing-holes on the perfor-

mances of the system. The experiments were per-

formed using the EKMB PIR sensors from Panasonic.

The data acquisition mechanism was implemented on

an nRF51-based hardware platform manufactured by

Nordic Semiconductors featuring a low-power SoC

that embeds an ARM Cortex-M0 MCU, along with

a 2.4 GHz wireless transceiver.

The considered deployment area has a rectangular

shape of size 3.3 × 2.4 m

. Most of occupants activ-

ity is concentrated over the ofﬁce desk that received

greater weights in the matrix Φ. The discretization

step l was ﬁxed to 0.3 m resulting in a grid of 11 × 8

points.

We have tested three deployments scenarios. The

ﬁrst one corresponds to the optimal solution of the

MPC problem when using one PIR. As shown in Fig.

2(a), this deployment covers nearly 63% of the desk’s

area. Optimal full coverage of this space is ensured

with 3 PIRs, which corresponds to our second deploy-

ment scenario depicted in Fig. 2(b). In the third sce-

nario, a single PIR was placed in a way to put the

largest holes at the desk area as shown in Fig. 2(c).

It shows the real impact of sensing-holes on the per-

formances of the detection system. It is worth noting

that existing solutions, by ignoring the presence of the

sensing-holes, consider such deployment as optimal

since the overall sensing range of a single PIR fully

covers the ofﬁce area.

The deployed motes actively monitor the state of

the PIR and notify a central base station about any

detection event. The latter maintains a database for

logging the incoming sensory data along with ground

truth presence/absence intervals, which are provided

manually by occupants. The experiments were per-

formed over a period of three days. The obtained re-

sults are depicted in Fig. 3(a) that summarizes, for

the three deployment scenarios, the proportions of all

possible detection cases: true presence (TP), true ab-

sence (TA), false presence (FP), and false absence

(FA). We can clearly see that taking into considera-

tion the presence of sensing-holes helps in reducing

the FA, i.e., the system is able to capture more occu-

pant movements. However, these results represent the

distribution of the raw data collected from PIRs and

can not be used as a reliable indication of absence.

As the PIR signal ﬂuctuates signiﬁcantly when occu-

pants are moving, detection systems generally imple-

ment a ﬁltering mechanism to smooth the collected

raw data. The ﬁlter is based on a timeout mechanism

that is launched when no motion is detected, which

delays the decision about absence detection to over-

come FA.

To evaluate the performance of the system in the

different deployment scenarios and under different

timeout values, we have measured two metrics, (i) the

comfort level, and (ii) the waste in energy usage. The

ﬁrst metric quantiﬁes the ability of the system to pre-

serve the convenience of users, that is, the ability not

to disturb the occupants by keeping ofﬁce energy sup-

ply on when they are present in the target area (i.e.,

ability to overcome FA). The second one reﬂects the

proportion of time the system fails to effectively de-

tect (or react to) the absence of occupants, which im-

plies a missed opportunity to reduce the energy con-

sumption.

Formally, the comfort level C and the energy us-

age waste W are computed as follows:

C =

+ F

, W =

+ T

where T

(respectively F

) denotes the total du-

rations of TP (respectively FP), and T

(respectively

) denotes the total durations of TA (respectively

FA).

For every deployment scenario, the value of the

absence timeout has been varied, and C , W have been

measured for every case. Fig. 3(b) shows the varia-

tion of the observed usage waste for different levels

of comfort. We observe that the performances of the

hole-unaware deployment are remarkably lower than

our proposed solution, which means that the presence

of holes signiﬁcantly affects the waste of energy us-

age, specially when requiring a high level of users’

comfort. In fact, to ensure a high level of comfort in

the presence of sensing-holes, absence decisions need

to be delayed for long periods (high timeout). This

is explained by the fact that these zones hamper the

proper capture of small movements, which increases

the time required to catch such rare events. The con-

sequence of high values of the timeout is that occu-

pants leaving the ofﬁce are not timely detected (re-

ported), which causes energy waste.

We can also notice from Fig. 3(b) that the perfor-

mances of the optimal solution using only one PIR are

SENSORNETS 2016 - 5th International Conference on Sensor Networks

178

Optimal 1 Optimal 3 Hole-unaware

10.75%

16.47%

4.6%

67.35%

67.33%

0.26%

0.28%

21.62%

15.9%

27.77%

True +

True -

False +

False -

(a)

2.5

7.5

50 60 70 80 90 100

Energy usage waste (%)

Comfort level (%)

Optimal 1

Optimal 3

Hole-unaware

(b)

Figure 3: Experimental results for the three scenarios. (a) Time proportion for the different cases of detection based on raw

data. (b) The variation of the energy waste for different levels of users’ comfort using a timeout smoothing.

very close to the optimal full coverage solution using

three PIRs. This is due to the fact the ﬁrst deploy-

ment covers an important proportion of the chair-side

of the ofﬁce desk. Consequently, covering the remain-

ing spaces of the desk, as in the second deployment

scenario, does not help capturing more occupants ac-

tivity. This result demonstrates that it is important

to properly construct the Φ matrix in order to focus

the optimization problem on the most relevant spaces

which helps reducing the number of required sensors.

4 CONCLUSIONS

In this paper, we have investigated the impact of

sensing-holes on the performances of typical PIR-

based occupancy detection systems. To our knowl-

edge, our work represents the ﬁrst study that (i) inves-

tigates how this intrinsic property affects the accuracy

of detection and (ii) proposes a deployment method to

alleviate its negative effects and enable the system to

optimize energy usage. The problem has been formu-

lated with mixed integer linear programming (MILP),

where the positions of a set of PIRs are sought out in a

way to maximize the effectively covered area. Based

on this formulation, several experiments have been

carried out to evaluate the performances of the ob-

tained solutions in comparison with hole-unaware de-

ployment. Results demonstrate clear improvements in

terms of accuracy in detection when using hole-aware

placement, which helps rationalizing energy manage-

ment. For example, in scenarios requiring high levels

of user comfort, a hole-unaware deployment can re-

sult in a 9.61% of waste of energy usage, while hole-

aware placement reduced this wastage to 1.3%.

The preliminary study presented in this paper is

the ﬁrst step in our ongoing project. The next one is

to use these results in the design of a wireless sensor

network for efﬁcient energy control in smart build-

ings. We think such an optimal deployment will help

to make the whole system effective, and to rational-

ize the settings of actuation parameters in a way that

balances user’s comfort and energy saving.

REFERENCES

Agarwal, Y., Balaji, B., Gupta, R., Lyles, J., Wei, M., and

Weng, T. (2010). Occupancy-driven energy manage-

ment for smart building automation. In Proc. the 2nd

ACM Workshop on Embedded Sensing Systems for

Energy-Efﬁciency in Building, BuildSys’10, pages 1–

Beltran, A., Erickson, V. L., and Cerpa, A. E. (2013).

Thermosense: Occupancy thermal based sensing for

hvac control. In Proc. of the 5th ACM Workshop

on Embedded Systems For Energy-Efﬁcient Buildings,

BuildSys’13, pages 11:1–11:8.

California Energy Commision (1993). Advanced Lighting

Guidelines.

Delaney, D. T., O’Hare, G. M. P., and Ruzzelli, A. G.

(2009). Evaluation of energy-efﬁciency in lighting

systems using sensor networks. In Proc. of the 1st

ACM Workshop on Embedded Sensing Systems for

Energy-Efﬁciency in Buildings, BuildSys’09, pages

61–66.

Kazmi, A. H., O’grady, M. J., Delaney, D. T., Ruzzelli,

A. G., and O’hare, G. M. P. (2014). A review

of wireless-sensor-network-enabled building energy

management systems. ACM Trans. Sen. Netw.,

10(4):66:1–66:43.

Lu, J., Sookoor, T., Srinivasan, V., Gao, G., Holben, B.,

Stankovic, J., Field, E., and Whitehouse, K. (2010).

The smart thermostat: Using occupancy sensors to

save energy in homes. In Proc. of the 8th ACM Confer-

ence on Embedded Networked Sensor Systems, Sen-

Sys’10, pages 211–224.

On the Effect of Sensing-holes in PIR-based Occupancy Detection Systems

179

Marchiori, A. and Han, Q. (2010). Distributed wireless con-

trol for building energy management. In Proc. of the

2nd ACM Workshop on Embedded Sensing Systems

for Energy-Efﬁciency in Building, BuildSys’10, pages

37–42.

Nguyen, T. A. and Aiello, M. (2013). Energy intelligent

buildings based on user activity: A survey. Energy

and Buildings, 56:244 – 257.

Panasonic (2012). Motion sensors datasheet.

http://www3.panasonic.biz/ac/e download/control/

sensor/human/catalog/bltn eng pir.pdf.

SENSORNETS 2016 - 5th International Conference on Sensor Networks

180