Directed Effort

A Generic Measurand for Higher Level Behavior Analysis

Benedikt Gollan

and Alois Ferscha

Pervasive Computing Applications, Research Studios Austria, Thurngasse 8/20, Vienna, Austria

Institute for Pervasive Computing, Johannes Kepler University, Linz, Austria

Keywords:

Automatic Behavior Analysis, Movement Tracking, Pattern Recognition, Estimation of Engagement.

Abstract:

Behavior and body language are essential components of human interaction. In this paper, we propose a

meta-level representation of human behavior for interpretative, higher level applications in human-computer

interaction systems called Directed Effort. A theoretical framework is described which is derived from behav-

ioral and psychological sciences and which is designed to represent the commitment and interest of people

towards objects via behavior analysis in real-life scenarios. Directed Effort, a score which allows the inter-

pretation of detected behavior changes is introduced as a generic measurand. Furthermore, a prototypical

implementation is documented to show the potential of the computed meta-level description of behavior.

1 INTRODUCTION

Behavior and body language are crucial aspects of

any kind of interaction between humans. Besides the

pure entropy of information which can be represented

bit-wise, it is the subtle indicators like the sarcastic

tone of voice, the rolling of the eyes or the impa-

tient tapping of the foot through which we transmit

the majority of the information which is necessary to

successfully interpret messages. Human interaction

is largely based on the meticulous interpretation of

meta-information for which, even among humans, ex-

perts are rare. This poses an immense challenge to all

researchers and designers of human-computer inter-

action systems that aim at creating natural and intu-

itive user interfaces, trying to best possible generate a

human-like interaction quality experience in human-

computer interaction systems.

Human behavior encompasses any observable

physical activity. Yet, The description of behavior

not only has to cover different aspects of behavior

depending on the respective ﬁeld of application, but

can also imply many different layers of abstraction.

These can range from a technical analysis of body

pose based on joint coordinates and orientations to a

higher-level interpretation of body language, or from

the pure analysis of movement speed data to an in-

terpretation of underlying motivations and intentions.

An ideal representation of behavior will of course in-

clude all behavioral aspects and layers, whereas ac-

tual applications and implementations will set limita-

tions to what is necessary, useful and technically fea-

sible.

As human interaction is largely based on the inter-

pretation of human behavior, we need to create repre-

sentations of behavior which allow further interpre-

tations of intentions that go beyond explicit user in-

put. In this work, we try to approach such a poten-

tial higher level description which may be applica-

ble to represent human commitment and interest. For

this purpose, we propose a generic meta-level descrip-

tion of behavior which is designed to be generally

valid, generic, independent from sensor technology,

and quantiﬁable to be suitable for various human-

computer interaction applications.

1.1 Related Work

Commitment and behavior have often been ap-

proached from different scientiﬁc ﬁelds which re-

sulted in numerous strategies towards speciﬁc aspects

and effects related to behavior, commitment and at-

tention. Elaborate surveys on human behavior analy-

sis have been composed by Ji et. al (Ji and Liu, 2010)

and Candamo et. al (Candamo et al., 2010).

Concentrating on technical realizations, the im-

plementations can be divided into the already in-

dicated categories of action or acitivity recognition

and a semantic representation of behavior. Activity

analysis can be covered by template-based or state-

Gollan B. and Ferscha A..

Directed Effort - A Generic Measurand for Higher Level Behavior Analysis.

DOI: 10.5220/0004696500830090

In Proceedings of the International Conference on Physiological Computing Systems (PhyCS-2014), pages 83-90

ISBN: 978-989-758-006-2

 2014 SCITEPRESS (Science and Technology Publications, Lda.)

space-based approaches. Employing templates, Bo-

bick and Davis (Bobick and Davis, 2001) ﬁrst used

motion energy images and motion history images to

represent and categorize activities. Blank et. al

(Blank et al., 2005) extracted the 3D silhouette of

people to enable a fast and robust classiﬁcation of

activities. State-space approaches interpret human

behavior as a state machine with different postures

as states. These are often represented via Hidden-

Markov-Models (HMMs) (Peursum et al., 2007), (Shi

et al., 2006).

Semantic descriptions of behavior have gained

momentum lately. The importance of body orienta-

tion in human interaction is investigated by Hieta-

nen (Hietanen, 2002) with the result that head and

upper body orientation are vital components which

nonverbally transfer the aspect of engagement in an

interaction. Guevara and Umemuro (Guevara and

Umemuro, 2010) implemented a behavior analysis to

infer on psychological states, ﬁnding that walking ve-

locity and motion load are suitable for predicting sad-

ness or neutral states of mind.

Usually, in human-computer interaction human-

computer interaction research, commitment is inter-

preted as equivalent to visual focus and no further

attention model is integrated. Smith et al. (Smith

et al., 2006) created a head tracking and gaze esti-

mation system which detects whether passing people

are actually watching a shop window. Yakiyama et

al. (Yakiyama et al., 2009) estimate commitment lev-

els towards target objects via a laser sensor and on

the basis of computed distance, basic orientation and

movement speed. According to Knudsen (Knudsen,

2007) orienting movements are used to optimize the

resolution of (visual) information about the object.

2 THE CREATION AND

CONTROL OF BEHAVIOR

Human behavior addresses any observable physical

activity of the individual. To enable a successful in-

terpretation, it is essential to understand basic under-

lying behavioral mechanisms. This chapter will de-

scribe a model derived from psychology and behavior

research of how behavior is motivated, controlled and

initialized.

Following Bongers (Bongers et al., 2009) and Di-

jksterhuis (Dijksterhuis and Aarts, 2010), behavior is

controlled via a sequential motivation chain (ﬁg. 1):

Input stimuli are processed and ﬁltered regarding their

value, importance and salience. The successive stim-

uli compete with already existing Motivations for

their realization. The result of this competition is a

set of prioritized Goals, that describe the intrinsic, of-

ten unconscious intentions of a person. These cannot

be assessed from the outside and may be as simple as

’being hungry’, or complex structures which can not

be verbalized at all. To actually achieve these abstract

Goals, we make concrete Plans like navigating to the

next restaurant, to satisfy the underlying motivation.

Finally, the actual realization of these plans leads to

the execution of observable physical behavior.

With outward behavior visualizing and represent-

ing inner states, a thorough analysis of pyhsical be-

havior holds great potential for a suitable analyis of

the level and even orientation of inner commitment.

Elementary changes in the motivation chain, e.g. trig-

gered by sudden extrinsic stimuli (siren, etc.) will

cause a sequence of re-prioritization of Goals and

Plans and ﬁnally result in alterations of physical be-

havior. Our approach is directed at the observation

and interpretation of such behavioral changes, to in-

fer alterations and repriorizations in the unfortunately

unobservable motivation chain.

The proposed model correlates very well with

existing behavior control ﬁndings. Posner (Posner,

1980) investigated the connection between extrinsic

behavior, e.g. head turning, eye movements towards

selected stimuli and inner processes which describe

all completely mental activities. Posner found the re-

lation between covert and overt attention to be not a

close but a functional one, showing a ’striking ten-

dency of attention to move to the target prior to an eye

movement’. Hoffman (Hoffman and Subramaniam,

1995) showed that, being ordered to direct gaze to-

wards a certain location, one cannot attend objects at

a different location. This existence of a neural struc-

tural connection between exterior and inner focus

was supported by Moore (Moore and Fallah, 2001),

Perry (Perry, 2000) and Rizzolatti (Rizzolatti et al.,

1987). On the other hand, experiments carried out by

Hunt and Kingstone (Hunt and Kingstone, 2003) in-

dicate that in case of bottom-up controlled, reﬂexive

processes overt and covert attention are strongly re-

lated whereas for top-down controlled processes, in-

ferring backwards from eye gaze alone to overt atten-

tion is error-prone. The difference between reﬂexive

and voluntary controlled mechanisms is supported by

uller and Rabbitt (M

uller and Rabbitt, 1989), who

detected higher reorientation performances for reﬂex-

ive reorientation of attentional focus.

PhyCS2014-InternationalConferenceonPhysiologicalComputingSystems

Figure 1: Schematic Illustration of the intrinsic and exterior processes of behavior control. (1) Incoming stimuli are ﬁltered

according to top down and bottom up processes (SEEV model (Wickens and McCarley, 2008)) (2) Succeeding stimuli enter

and alter the Motivation Chain and inﬂuence the distribution of Attention Resources (3) Realization of Plans expresses in

observable behavior (4) Behavior changes can be tracked, quantiﬁed and interpreted.

3 EFFORT AS KEY-PARAMETER

FOR BEHAVIOR ANALYSIS

Having a model of how behavior is generated, the next

step is to ﬁnd a suitable representation which is char-

acteristic for all kinds of behavior and especially de-

scribes changes of behavior in a qualitative and quan-

titative way.

Every alteration of existing plans and accordingly

of current behavior is characterized by its demand for

a certain amount of Mental Effort, which includes the

process of ﬁltering stimuli input and deciding to com-

mit to a source of information and consequently a

rescheduling of future tasks. Furthermore, it requires

Physical Effort, ‘an important concept, ... required to

access different sources of information, using what-

ever mechanism is necessary: eyes, head, body, hands

or even the walking feet’ (Wickens and McCarley,

2008), to actually alter physical behavior.

In this context, the principle of the economy of

movement represents a crucial aspect. As Bitgood

states: ‘To overcome the economy of movement moti-

vation,.. , the perceived beneﬁts of approaching an at-

tractive object must outweigh the perceived cost of the

effort’ (Bitgood, 2006). Generally, people tend to op-

timize their behavior concerning energy consumption

and effort, physical or mental, as ‘excessive mental ef-

fort, like excessive physical effort, generally produces

an unpleasant state that is to be avoided. Hence, peo-

ple tend to be inherently effort conserving, particu-

larly when placed in high demanding environments...’

(Wickens and McCarley, 2008). Consequently, we as-

sume that one will stick to his current comportment

until given valid reason to change, thus evading un-

necessary investment of effort.

To give an example, in spatial contexts, behavior

control is inﬂuenced by the effort which is necessary

to access information sources. This physical effort in-

volves all overt processes from eye movement, over

body posture to movement parameters. Kahneman

states that ’because of the connection between effort

and arousal, physiological measures of arousal can

be used to measure the exertion of effort’ (Kahneman,

1973). According to Knudsen (Knudsen, 2007) Ori-

enting Movements are used to optimize the resolution

of (visual) information about the object. To demon-

strate these principles, a sample scenario of a person

moving through a shopping mall is displayed in ﬁg-

ure 2. Passing a public display, there are different be-

haviors that can be adopted. In the sample scenario,

the described options (a)-(c) differ in the amount of

energy and time which are invested to engage with

the display and the presented content. As can be ob-

served, the more effort is invested the higher will be

the commitment to the object.

Bringing it all together as illustrated in ﬁgure 1, in

the complete process of (i) ﬁltering stimuli, over (ii)

alteration of the motivation chain to the (iii) allocation

DirectedEffort-AGenericMeasurandforHigherLevelBehaviorAnalysis

Figure 2: Different kinds of behavior in a mall scenario when passing a public display. (a, white) Passers-by may not perceive

the display at all and show no reaction, (b, yellow) turn their head towards the screen but continuing current general behavior

of approaching their destination or (c, green) actually changing the path, investing time and commitment in perceiving the

presented information.

of attention resources and ﬁnally to the (iv) execution

of related plans to satisfy underlying motivations, it

is named Effort which represents the critical thresh-

old of whether attention resources are assigned and

behavior is changed. At the same time it represents

an observable indicator of physical commitment to a

source of information. With effort being (not the only

but) such an important regulating factor in behavior

control, we propose physical effort as the basis for a

generic and generally applicable higher level repre-

sentation of behavior.

4 BEHAVIOR ANALYSIS

FRAMEWORK

Providing a valid measurement of behavior, demands

three important elements: (a) measuring directly sig-

niﬁcant target behavior, (b) measuring a relevant di-

mension of target behavior and (c) ensuring that the

data is representative for the given use-case (Cooper

et al., 2007). In this chapter, a behavior measure

framework is proposed based on interpreting behav-

ioral changes which tries to best possibly match these

prerequisites.

To approach an actual implementation, changes of

behavior need to be detected, quantiﬁed and evalu-

ated. We assume an interaction system with any kind

and number of sensor-based tracking system(s) (cam-

eras, distance sensors, depth sensors, etc.) with which

behavioral data can be collected. The collected be-

havior data depends on the choice of sensor and appli-

cation scenario and may include any measurable data

that describes movement in a characteristic way like

skeleton joint coordinates, gaze direction, etc.

Our proposed framework (ﬁg. 3) deﬁnes four im-

portant variables, which are behavioral parameters

(t), Effort e

(t), Effect f

(t) and Directed Effort

DE(t).

(t) = s(i,t) (1)

(t) = ∆(b

(t),B

;t −1))) (2)

(t) = Φ(b

(t),

→

) (3)

DE(t) =

∑

· e

(t) · f

(t) (4)

Behavior parameters b

(equ. 1) describe the sen-

sor data s(t) extracted to exclusively describe a single

aspect of behavior i. These could range from move-

ment speed, orientation or location of single body

joints to a emitted volume of whole groups of peo-

ple. The selection of these behavioral features heav-

ily depends on the application and choice of sensor.

For later evaluation of the data, it is necessary to best

possible isolate the characteristic parameters of the

aspired behavior parameter in the feature extraction

process. Please note, the behavior parameters may

range from mere numeric to directed dimensions like

vectors, angles, etc. Due to this variety, the notation

has been conﬁned to an abstract placeholder b

The extracted behavior data is used to calculate

changes in behavior individually for each behavior

parameter per frame t. For this purpose, training data

is collected to describe recent reference behavior

which is used to detect the amount of alterations. The

process of calculating the amount of effort e

(t) which

is required to execute the detected change of behavior

(equ. 2) is represented via ∆ (equ. 5). In this func-

tion, effort is represented via a percental represen-

PhyCS2014-InternationalConferenceonPhysiologicalComputingSystems

Figure 3: Visualization of behavior analysis framework components. Tracking data is collected from sensor(s). Feature

extraction algorithms provide behavior data which is used to calculate reference behavior and current behavior changes. The

changes of behavior are processed to effort scores and evaluated in relation to the location of the object of interest to obtain

effect information. Finally, effort and effect are combined to a single expressive feature called Directed Effort which describes

the level of engagement of behavior changes to an object of interest.

tation in relation to the maximal possible change of

behavior for the respective behavior parameter. E.g.,

turning the head with an angle of 20

◦

and assuming

a maximum turn of 180

◦

would result in an effort of

◦

/180

◦

= 11, 1% for the behavioral parameter of

head turn. In spite of the issue of deﬁning these maxi-

mum values, this process provides a normalized level

of effort scores throughout completely unrelated pa-

rameters of behavior, making them comparable.

∆(b

(t),B

; t − 1]) =

(t) − B

(t)

i,max

(5)

These effort scores e

represent the overall and un-

evaluated invested effort scores. Accumulating over i

would deliver the overall amount of invested effort.

Yet, the proposed isolated representation of the ef-

fort per behavior parameter is necessary for the fol-

lowing evaluation of the orientation and effect of the

invested effort. Assuming that all behavior changes

are causally determined, our approach tries to inter-

pret the effect of the detected behavior to deduce the

source of the behavioral change and thus draw con-

clusions on the underlying motivations.

Given the vast amount of potential objects of in-

terest and the high possibility that the actual sources

of the behavior change are unknown and outside of

the observable area, the evaluation has to be restricted

to distinct objects. This restriction will not allow a

general identiﬁcation of sources of interest but at least

enables the observation and evaluation of distinct, sin-

gle targets. By deﬁning the location

→

of an object of

interest (OOI) in the observed scene, it is possible to

analyze the effect f

of the change on potential com-

mitment to this OOI. The evaluation of the behavior

data (equ. 3) in relation to the location of a reference

OOI is represented by Φ (equ. 6). Again, this func-

tion needs to be accustomed to the speciﬁc behavior

parameter and a percental representation is proposed.

Φ(e

(t),

→

) =

→

(t,

→

)

i,opt

(6)

As a ﬁnal step, extracted information concerning

the amount and the effect of the invested Physical Ef-

fort are combined into a single expressive value called

Directed Effort (equ. 4) which represents the effective

invested Effort which is has been evaluated as con-

tributive regarding potential OOI. In other words, Di-

rected Effort scores describe how much effort which

is directed at a speciﬁc object, has been invested.

The already calculated values e

(t) and f

(t) hold the

amount and effect of the respective behavior parame-

ters i. To combine them to a single expressive score,

the products of the corresponding effort and effect

scores are accumulated and weighted with a factor

. This weighting again depends on the application

scenario and choice of parameters. By interpreting

these Direct Effort scores, we hope to be able to draw

conclusions on the distribution of attention resources

which has evoked the observed behavior.

5 EXEMPLARY

IMPLEMENTATION

To demonstrate the functionality of the proposed

framework, a sample implementation is described in

the following. A public display scenario has been se-

lected, in which the commitment of passers-by to the

displayed content is supposed to be investigated. To

enable behavior analysis, a large-scale public display

has been equipped with a depth sensor which allows

an accurate tracking of body pose and extraction of

movement features of passers-by.

First, the aspired behavior parameters b

need to

be identiﬁed and deﬁned. In the given scenario, be-

havior can be unraveled into the three existing degrees

DirectedEffort-AGenericMeasurandforHigherLevelBehaviorAnalysis

(a) (b)

Figure 4: (a) Visualization of behavior parameters b

, b

and (b) b

of freedom of movement which are movement direc-

tion, body orientation and velocity.

Body orientation is set identical to head orienta-

tion, as it turned out to be the orientation component

of the highest relevance and the best tracking results

in our implementation. The effort component which

derives from movement direction and body orienta-

tion is calculated as the difference of the detected

current angles b

and b

(ﬁg. 4) to learned current

reference values B

and B

. In this application, the

’learned’ components B

are implemented as an expo-

nentially decreasing low-pass ﬁlter (equ. 10). Finally

the alteration is set in relation to the maximum change

per parameter, which are ±180

◦

for movement direc-

tion and ±90

◦

for head orientation.

To calculate the effort invested in a change of

speed, acceleration values b

can be analyzed. Yet,

here a different approach is followed, since relation

to a theoretical maximal acceleration value does not

adequately describe real circumstances. In this case,

an acceleration from 5

to 10

would result in the

same effort as an increase from 15

to 20

, although

in the ﬁrst case the speed has been doubled. This is

why a different percentaged representation has been

chosen which describes the percentaged change in re-

lation to recent velocity. Note, that only effort from

behavior changes are analyzed at this point, although

of course, maintaining a high speed will involve im-

mense physical effort. The inclusion of these constant

aspects of effort are part of current research for the

generalization and enlargement of the framework.

(t) =

(t)

(7)

(t) =





(t) − B





(8)

(t) =

(t)

− B

180

(9)

∑

n=0

· b

(t − n) (10)

To analyze the effect of the invested effort towards

the display location, the orientation of the behavioral

changes has to be evaluated in relation to the loca-

tion of the display, which results in a measure of

the contributivity of the detected activities to the dis-

play location. E.g. a person making a turn of 90

◦

will always result in the same detected physical ef-

fort no matter where in the scene the activity took

place. But is the interpretation of the orientation to-

wards the display which elevates the detected physical

effort from a pure mathematical description of activ-

ity to an expressive representation of potential com-

mitment, which is Directed Effort. It enables us to in-

terpret whether activities are directed towards or away

from the display, bringing the person to a more or less

attentive state.

To evaluate the acceleration parameter b

(t) (ﬁg.

5(a)) a rule-based evaluation function φ

is selected.

First, the absolute effective fraction of the accelera-

tion vector

−→

a to the display location is analyzed to

evaluate to what degree the activity is related to the

display. Second, a rule-based evaluation of the con-

tributivity is applied which relies on the assumption

that any activity which increases the stay in the range

of the display or improves the perception of the dis-

play is considered as contributive. Hence, all decel-

eration and movement into the display sector are in-

terpreted as positive effect, whereas all accelerations

which are directed away from the display are evalu-

ated as negative effect (equ. 13).

(a) (b) (c)

Figure 5: Computation of the Effect of (a) the acceleration

∆b

−→

(t) by computing the component of f

−→

(t) that is di-

rected at the object (orthogonal projection of ∆b

−→

(t) onto

the direct connection to the object location), (b) b

(t) by the

change of movement angle towards the object and (c) b

(t)

by change of orientation angle f

(t) towards the object.

(t,

−→

x ) = ρ

(t) · b

v, f

(t) (11)

v,x

(t) =

−→

x •

−−→

(t)

−→

x (12)

(t) =

(

1, if b

(t) < 0

ω[t] ≤ ε

,ε

−1, else.

(13)

PhyCS2014-InternationalConferenceonPhysiologicalComputingSystems

Figure 6: Directed Effort Scores plotted over time with corresponding behavior illustration. The scene is divided into phases

(I-V). (I): Person walking parallel to display with head turned towards the screen. (II) and (III) Person stopping to watch,

movement direction still parallel and head turned. (IV) Person approaching the screen changing direction assuming a close and

comfortable position with head and shoulders oriented in the same direction. (V) Turning and leaving the scene. The dotted

curves show the result for the single components of Directed Effort which are aggregated to an overall score of Directed Effort

(solid line). Positive amplitudes express positive contributions, negative amplitudes indicate activities which are directed away

from the display.

Considering the effect of variations in movement

direction and orientation, changes are considered ben-

eﬁcial, if the movement causes the angle towards the

display location to decrease (equs. 14, 15). Again, the

effect scores are percentaged in relation to the maxi-

mal value.

(t) =

ϕ,x

− b

ϕ,x

(t))

180

(14)

(t) =

τ,x

− b

τ,x

(t))

180

(15)

DE =

∑

· e

(t) · f

(t) (16)

Bringing it all together, the effort and effect scores

are weighted with parameters α

and accumulated to

the ﬁnal Directed Effort score. The weighting param-

eters are necessary to achieve realistic, well-balanced

effort scores. They have to be established during a

test phase of the system for each application.

Referring to the stated requirement from chap-

ter 4, it can be summarized, that the framework and

the exemplary implementation can claim to fulﬁll the

speciﬁcations. The described behavior measurement

procedure directly measures signiﬁcant behavior in a

useful dimension and works on data from real life sce-

narios.

6 RESULTS

To evaluate the framework functionality, plotting DE

scores over time provides useful information. It re-

sults in effort curves which are useful to present the

applicability and the potential of our effort-based be-

havior analysis approach. The resulting curves suc-

ceed in adequately describing behavior and signal

changes of behavior via strong signal peaks. In ﬁgure

6, the effort curves have been plotted for the single DE

parameters acceleration, moving direction, and shoul-

der and head orientation which contribute to the over-

all Directed Effort curve, using an example from a

database which was gathered during an installation at

a public event. This sample has been divided into ﬁve

sections which mark different behavioral segments.

As can be observed, the calculated DE curve

gives a suitable expression of invested physical ef-

fort and enables detection and interpretation of be-

havioral changes. The beginning of each of the sec-

tions which indicate different kinds of activities is

marked with a strong peak in the DE curve, indicat-

ing a high effort level and a substantial change in

behavior as demanded in our theoretical approach.

The height and width of the peak represent an in-

dicator for the strength of the behavior modiﬁcation

and the algebraic sign clearly separates contributive

from detrimental actions. Smaller behavioral changes

like the deceleration in sections II and III result in a

lower peaks than changes which include strong devi-

ations in in more than one of the single parameters,

as in section IV and V. In the latter sections, the per-

son not only alters movement speed, but as well the

movement direction and body orientation resulting in

higher DE scores.

The noise-like patterns which occur throughout

the sample are mainly caused by oscillations in the ac-

DirectedEffort-AGenericMeasurandforHigherLevelBehaviorAnalysis

celeration parameter, which derive from the gait fre-

quency and the slightly strolling walking style of the

subject. Overall, DE curves show promising stability

and at the same time reactivity to behavior modiﬁca-

tions and seem to adequately describe the qualitative

commitment of people towards objects like public dis-

plays.

7 CONCLUSIONS

AND OUTLOOK

In this paper, we have presented an approach towards

a higher level interpretative description of behavior

to express engagement and commitment of via detec-

tion of behavior changes. Such an approach can never

claim to be able to predict the exact focus of atten-

tion of a person but can only try to provide a model

which approximates reality through iterative reﬁne-

ment. The more we accomplish a detailed description

of behavior and context, the better we will perform in

interpreting human behavior. Yet, the proposed meth-

ods may provide a ﬁrst step towards a behavior-based

attention estimation.

REFERENCES

Bitgood, S. (2006). Not another step! economy of move-

ment and pedestrian choice point behavior in shopping

malls. Environment And Behavior, 38(3):394–405.

Blank, M., Gorelick, L., Shechtman, E., Irani, M., and

Basri, R. (2005). Actions as space-time shapes. In

Computer Vision, 2005. ICCV 2005. Tenth IEEE Inter-

national Conference on, volume 2, pages 1395–1402

Vol. 2.

Bobick, A. and Davis, J. (2001). The recognition of human

movement using temporal templates. Pattern Analy-

sis and Machine Intelligence, IEEE Transactions on,

23(3):257–267.

Bongers, K. C., Dijksterhuis, A., and Spears, R. (2009).

Self-esteem regulation after success and failure to at-

tain unconsciously activated goals. Journal of Exper-

imental Social Psychology, 45(3):468 – 477.

Candamo, J., Shreve, M., Goldgof, D., Sapper, D., and Kas-

turi, R. (2010). Understanding transit scenes: A sur-

vey on human behavior-recognition algorithms. Intel-

ligent Transportation Systems, IEEE Transactions on,

11(1):206–224.

Cooper, J. O., Heron, T. E., and Heward, W. L. (2007).

Applied behavior analysis. Pearson/Merrill-Prentice

Hall, Upper Saddle River and N.J, 2 edition.

Dijksterhuis, A. and Aarts, H. (2010). Goals, Attention,

and (Un)Consciousness. Annual Review of Psychol-

ogy, 61(1).

Guevara, J. E. and Umemuro, H. (2010). Unobtrusive esti-

mation of psychological states based on human move-

ment observation. e-Minds, 2(6).

Hietanen, J. (2002). Social attention orienting integrates vi-

sual information from head and body orientation. Psy-

chological Research, 66(3):174–179.

Hoffman, J. E. and Subramaniam, B. (1995). The role of vi-

sual attention in saccadic eye movements. Perception

& psychophysics, 57(6):787–795.

Hunt, A. R. and Kingstone, A. (2003). Covert and overt

voluntary attention: linked or independent? Brain

research. Cognitive brain research, 18(1):102–105.

Ji, X. and Liu, H. (2010). Advances in view-invariant hu-

man motion analysis: A review. Systems, Man, and

Cybernetics, Part C: Applications and Reviews, IEEE

Transactions on, 40(1):13–24.

Kahneman, D. (1973). Attention and effort. Prentice-Hall,

Englewood Cliffs, N.J.

Knudsen, E. I. (2007). Fundamental components of atten-

tion. Annual Review of Neuroscience, 30(1):57–78.

Moore, T. and Fallah, M. (2001). Control of eye move-

ments and spatial attention. Proceedings of the Na-

tional Academy of Sciences of the United States of

America, 98(3):1273–1276.

uller, H. J. and Rabbitt, P. M. (1989). Reﬂexive and vol-

untary orienting of visual attention: time course of

activation and resistance to interruption. Journal of

experimental psychology. Human perception and per-

formance, 15(2):315–330.

Perry, R. J. (2000). The neurology of saccades and covert

shifts in spatial attention: An event-related fMRI

study. Brain, 123(11):2273–2288.

Peursum, P., Venkatesh, S., and West, G. (2007). Tracking-

as-recognition for articulated full-body human motion

analysis. In Computer Vision and Pattern Recognition,

2007. CVPR ’07. IEEE Conference on, pages 1–8.

Posner, M. I. (1980). Orienting of attention. Quarterly Jour-

nal of Experimental Psychology, 32(1):3–25.

Rizzolatti, G., Riggio, L., Dascola, I., and Umilt

a, C.

(1987). Reorienting attention across the horizontal

and vertical meridians: evidence in favor of a premo-

tor theory of attention. Neuropsychologia, 25(1A):31–

40.

Shi, Y., Bobick, A., and Essa, I. (2006). Learning tem-

poral sequence model from partially labeled data. In

Computer Vision and Pattern Recognition, 2006 IEEE

Computer Society Conference on, volume 2, pages

1631–1638.

Smith, K. C., Ba, S. O., Odobez, J.-M., and Gatica-Perez, D.

(2006). Tracking attention for multiple people: Wan-

dering visual focus of attention estimation. Idiap-RR

Idiap-RR-40-2006, IDIAP. Submitted for publication.

Wickens, C. D. and McCarley, J. S. (2008). Applied atten-

tion theory. CRC Press, Boca Raton.

Yakiyama, Y., Thepvilojanapong, N., Iwai, M., Mihirogi,

O., Umeda, K., and Tobe, Y. (2009). Observing real-

world attention by a laser scanner. IPSJ Online Trans-

actions, 2:93–106.

PhyCS2014-InternationalConferenceonPhysiologicalComputingSystems