Different Intelligent Approaches for Modeling the

Style of Car Driving

Jose Aguilar

1,2,3

, Kristell Aguilar

, Danilo Chávez

, Jorge Cordero

and Eduard Puerto

CEMISID, Universidad de Los Andes, Mérida, Mérida, Venezuela

Escuela Politécnica Nacional, Pichincha, Quito, Ecuador

DCCE, Universidad Técnica Particular de Loja, Loja, Ecuador

GIDIS, Universidad Francisco de Paula Santander, Cúcuta, Colombia

Keywords: Hierarchical Patterns, Fuzzy Logic, Chronicles, Dynamic Pattern Recognition, Style of Driving.

Abstract: In this paper, we propose a hierarchical pattern of the style of driving, which is composed of three levels, one

to recognize the emotional state, other to recognize the state of the driver, and finally, the last one corresponds

to the style of driving. Each level is defined by different types of descriptors, which are perceived in different

multi-modal ways (sound, vision, etc.). Additionally, we analyze three techniques to recognize the style of

driving, using our hierarchical pattern, one based on fuzzy logic, another based on chronicles (a temporal

logic paradigm), and another based on an algorithm that models the functioning of the human neocortex,

exploiting the idea of recursivity and learning in the recognition process. We compare the techniques

considering the dynamic context where a car driver operates.

1 INTRODUCTION

With the popularity of advanced systems driver

assistance (ADAS) in vehicles, and setting the

context of a man-machine system, the problem of

interaction between drivers and ADAS becomes

important, but more important is how adapts it to the

characteristics of each driver.

In order to make the ADAS can suit to individual

drivers, it is necessary that the ADAS can count

adaptive systems that can consider internal

characteristics of each human being, as fatigue,

inattention, and in this case, its type of driving (Lin et

al. 2014). There are a lot of work about the emotions

in a car, e.g., in (Aguilar et al. 2016), (Cordero &

Aguilar 2016) is proposed a recognition model of the

emotional state, using chronicles and static patterns.

On the other hand, in (Eyben et al. 2010) show how

the emotions are a key issue not only in a general

oncoming human-computer interaction, but also in

the in-car communication. (Katsis et al. 2015) present

a revision of the works in emotion recognition,

focusing on those influencing the driver's

performance. The work of (Aypar et al. 2014) is

focused on an alerting mechanism based on the driver

state recognition. (Guoying & Danpan 2016) propose

a pattern recognition approach to identify the driver

steering behavior. There are much more works about

the emotions of the car driver, but in general, they

propose simple models, or they study only the

emotions (Kolli et al. 2011), (Tawari & Trivedi

2010), (Paschero et al. 2012), (

Wang, J. et al., 2013).

The main contribution of this paper is to propose

a hierarchical pattern of the style of driving, which

consider three levels of recognition, one to recognize

the emotional state, other to recognize the state of the

driver, and finally, the last one corresponds to the

style of driving. Each level is composed of different

descriptors, which require a multi-modal approach in

order to be perceived, and they are related between

them because they are descriptors between them. In

addition, the paper analyses three techniques to

recognize the style of driving, one based on fuzzy

logic, another based on chronicles, and another based

on an algorithm that models the functioning of the

human neocortex, called Ar2P. We compare the

techniques, evaluating their capabilities to define

countersteering strategies, to adapt it to the driver, or

to Internet of things (IoT).

284

Aguilar, J., Aguilar, K., Chávez, D., Cordero, J. and Puerto, E.

Different Intelligent Approaches for Modeling the Style of Car Driving.

DOI: 10.5220/0006411902840291

In Proceedings of the 14th International Conference on Informatics in Control, Automation and Robotics (ICINCO 2017) - Volume 2, pages 284-291

ISBN: Not Available

2 FORMAL DEFINITION OF THE

PATTERN OF THE STYLE OF

DRIVING OF A CAR DRIVER

In general, a pattern can be considered as the

abstraction of a set of objects, and normally is defined

by a set of descriptors. In this paper, we propose to

model the style of driving of the car driver using a

hierarchical pattern, which is composed of 3 levels:

First level: Pattern of the style of driving. Its aim

is to model how the driver drives. In the literature,

classically the style of driving can be aggressive,

ecological, urban, and normal. This level must detect

the style, based on the descriptors of the Table 1.

Table 1: Descriptors of the Pattern of the style of driving.

Descriptor Description

Type of roads It describes the type of the road.

Driver state

It describes the state of the car driver,

and it is defined by the second level of

our pattern.

Emotion of the

driver

It defines the emotional state of the

driver, and it is defined by the third

level of our pattern.

Environmental

condition

It characterizes the current

environmental conditions.

States of the

road

It characterizes the current conditions of

the road.

Traffic

characteristic

It defines aspects linked to the transit

laws.

Second level: Driver state. Its aim is to describe

the state of the car driver. In the literature, normally,

the state of a car driver can be wakeful, stressed,

lethargic, pleasant, fatigued, calm, boring, falling

asleep, among others. This level must detect the state

of the driver, based on the descriptors of the Table 2.

Table 2: Descriptors of the Pattern of the driver state.

Descriptor Description

Class of the

vehicle

It describes the type of vehicle.

Action control

over the vehicle

It describes the current action of the

driver of the car.

Emotion of the

driver

See description in Table 1.

Vehicle

condition

It defines the current conditions of the

vehicle.

Characteristics

of the driver

It defines the profile of age, or physical

condition, of the driver.

Driving

experience

It characterizes the experience of the

driver as a car driver.

Driving hour It defines the current hour of the day.

Third level: Emotions of the Driver. Its aim is to

describe the emotions of the driver. This level must

detect the current emotion of the car driver.

Particularly, we are going to use the six basic

emotions defined in the literature: happiness, sadness,

fear, anger, disgust, and surprise. The descriptors that

define this pattern are described in Table 3.

Table 3: Descriptors of the Pattern of the Emotions of the

Driver.

Descriptor Description

Driver behavior

It defines the current behavior of the

driver in the vehicle.

Action control

over the vehicle

See description in Table 2.

Physiological

behavior of the

driver

It defines the current physiological

conditions of the driver.

Vehicle condition See description in Table 2

Voice

expressions of

the driver

It characterizes the current tone of

voice of the car driver.

Facial

expressions of

the driver

It characterizes the current facial

expressions of the car driver.

Body expressions

of the driver

It describes the current body

expression of the driver.

The main goal of the hierarchical pattern is to

recognize the style of driving. To recognize the

style of driving, we need different descriptors (see

Table 3), which describe it. Particularly, one of the

descriptors is the state of the driver, which again is

described by a set of descriptors (see Table 2). Other

descriptor is the emotional state of the driver, which

also is described by a set of descriptors (see Table 1).

Thus, each level has a different set of descriptors,

which are perceived in different ways (sound, vision,

etc.) that implies to use a multi-modal approach for

the perception.

The descriptors describe various aspects: facial,

acoustic, body language, among others. The current

status of the descriptors are determined by the events

that are captured in the environment of the vehicle in

a given moment. For that, we use information from

the different sensors in the car, to characterize these

events. For example, for the speed of the car, we can

define the set of events of the Table 4. And so for the

rest of descriptors of our hierarchical multimodal

model.

Now, according to the current values of the

descriptors, are determined the current emotion of the

driver, the current state of the driver, and finally,

his/her style of driving, using the hierarchical

Different Intelligent Approaches for Modeling the Style of Car Driving

285

multimodal model. Table 5 shows an example of the

possible emotions recognized by the pattern of the

third level of our hierarchical multimodal model,

according to the value of the descriptors of this

pattern. Table 6 shows an example of the possible

style of driving recognized by the pattern of the first

level of our hierarchical multimodal model, according

to the value of the descriptors of this pattern. It is

important to remark that Table 5 and 6 show some of

the emotions and styles that can be recognized, Also,

they show some of the possible combinations of the

values of the descriptors for the recognition of these

emotions and styles (e.g., Table 5 shows two

examples of events (ED2 and ED3) to recognize the

"happiness" emotion, but there may be more

combinations of values of the descriptors to recognize

it). For the possible states of the driver (second level

of our hierarchical multimodal model), it is similar.

Table 4: Events about the speed of the car.

Id Event Description Speed

S1 High speed > 100 Km/h

S2 Normal speed ≥ 40 and ≤ 100 Km/h

S3 Low speed < 40 Km/h

3 APPROACHES FOR THE

MODELING OF THE STATES

OF A CAR DRIVER

3.1 Based on Chronicles

A chronicle can be defined as a set of events, linked

by a set of temporal constraints (Aguilar 2011). Each

chronicle is an event pattern with temporal

relationships between them, and a set of chronicles

characterizes the possible evolution of a system

studied. To define a chronicle, normally two

predicates are used: event and hold. An event

expresses a change in an attribute, for example:

Event(state(light): (on, off), t2). A hold specifies that

an attribute holds a value during a time interval, for

example: Hold(position(robot, home), (t2, t4)).

In general, a chronicle model C is defined by a

pair (S, T), where S is the set of events and T the

temporal constraints between the events. A chronicle

instance c of a chronicle model C is a set of event

occurrences, which is consistent with the time

constraints of C.

The hierarchical pattern recognition system based

on chronicles paradigm consists of 3 types of

chronicles: i) First type, represents the emotional

patterns of the driver. Its aim is to describe the

emotions of the driver; ii) Second type, represents the

patterns of the driver state. Its aim is to describe the

driver's condition; iii) Third type, represents the

patterns of the driving styles. Its aim is to establish

how the person drives.

Every emotion, state, or driving style of the driver

will be modelled by a different chronicle, which

contains the events and the temporal relationships to

recognize them. A specific emotion, state or driving

style can be recognized by several chronicles, an each

chronicle is defined by the set of descriptors defined

in the previous section. An example of a chronicle of

the first type, to recognize the anger, is:

Chronicle Anger {

event(F3, T4),

event(P1, T3),

event(B5, T5),

event(H1, T6 ),

event(V1, T1),

event(S1, T2)

T1 T3,

hold(F3, (4, 10)),

hold(S1,(6, 20)),

When recognized {emit event(ED1)}}

According to this chronicle, the pattern of anger

can be recognized when the voice event "Tone treble

and volume high and speaking rate fast" (V1) arrives

at time T1, and holds between 4 and 10 units of time;

the speed event "High speed" (S1) occurs at the time

T2, and holds between 6 and 20 units of time, the

pressure event "Strong pressure of the steering wheel"

(P1) appears at time T3 and it is less than or equal to

T1, the facial event " Eyes and Eyebrows open, with

curves and tight lips, and face wrinkles in the center"

(F3) ocurrs at time T4, the body event "Posture

Flattened" (B5) ocurrs at time T5, and the heart event

"Fast Heast rate" (H1) arrives at time T6.

An example of a chronicle of the third type, to

recognize an aggressive driver, is the following:

Chronicle Aggressive {

event(ED1, TED1),

event(ST3, TST3),

event(R1, T4),

event(E2, T3)

TED1 TST3,

T3 3 T4,

hold(ST3, (5, 15)),

When recognized {Report the style of driving to

the driver assistance system}}

ICINCO 2017 - 14th International Conference on Informatics in Control, Automation and Robotics

286

Table 5: Emotions of a driver.

Event

Emotion Driver behavior Action

control over

the vehicle

Physiological

behavior of the

driver

Vehicle

condition

Voice

expressions of

the driver

Facial

expressions of

the driver

Body

expressions of

the driver

ED1 anger the car driver

pulls the door

pressing the

steering-

wheel

heart rate high,

the pupil

dilatation of the

driver high

mechanical

failure or

electrical

failure

the driver is

shouting

the driver is

serious

the driver

moves

violently

ED2 happiness the driver uses

the seat belt

normal heart rate normal normal the driver is

singing

the driver is

smiling

the driver

reacts slowly

ED3 happiness the driver uses

the seat belt

the driver is

calm

normal heart rate normal normal the driver is

whistling

the driver has a

calm face

the body of the

driver is calm

ED4 fear the driver uses

the seat belt

braking the color of the

face white

any normal the driver is

serious

Table 6: Style of driving.

Event

State of the

driver

Type of

roads

Driver

state

Emotion of

the driver

Environmental

condition

States of the road Traffic characteristic

SD1 aggressive any stressed anger is raining the road has potholes does not follows traffic signs

SD2 ecological rural relaxed happiness any any follows speed limits

SD3 normal urban relaxed happiness any any any

The structure of the chronicles of the second is

similar. This is just a sample of the proposed

chronicles used by the ADAS, where: i) The emotions

(anger, happiness, fear, among others) make up the

chronicles of type 1 (EDi), representing the emotional

patterns of the driver; ii) The driver states (stress,

pleasant, wakefulness, sleepy, among others) are the

chronicles of type 2 (STi), representing the patterns

of driver states; iii) The styles of driving (aggressive,

ecological, normal) are the chronicles of type 3 (SDi),

representing the patterns of the driving styles.

The chronicles of type 1 and 2 are composed of

the primary events captured through different types of

sensors (pressure sensor on the steering wheel,

driver's heart rate sensor, speed sensor, among

others). The chronicles type 3 are a mixture of the

primary events and the events recognized in the

hierarchical system. This level communicates with

the driving system to generate the relevant actions

according to the identified driving style.

3.2 Based on Ar2P

Ar2P (Algoritmo Recursivo de Reconocimiento de

Patrones, for its acronym in Spanish) is a model for

pattern recognition, inspired in the pattern recognition

theory of mind (Puerto & Aguilar, 2016), (

Puerto &

Aguilar, 2017). Each layer in the hierarchy is an

interpretation space identified as Xi, from i=1 to m.

X1 is the level of recognition of atomic patterns, and

Xm is the level of recognition of complex patterns (a

complex pattern is characterized by being composed

of patterns of lower levels). Each level is composed

of Γji recognition modules, (for j = 1, 2, 3... # of

modules at level i). ρji is the recognized pattern by the

module j at level i. The function of each recognition

module is to recognize its corresponding pattern. s()

represents the presence of a pattern to be recognized.

This input is specific to each recognition module. For

the top-down case, the output signal of the higher-

levels is the input signal at the lower-levels.

There is a ν relationship of structural composition

among the Γji of different Xi, such that Γrt → Γlk,

where t < k, and the relationship “→“ indicates that

Γrt of Xt is contained or forms part of Γlk, which

belongs to layer Xk of higher level. There may be

different versions of the same pattern

(redundancy/robustness) represented by different Γrt,

from r = 1,2,3...until possible variations of the object

in the real world. Each level i produce an output

signal (recognition or learning) based on the

responses of its modules. The output of each Γji

consists of a specific signal of recognition of its

pattern ρji, which is transmitted through the dendrites

to its higher levels. This signal contains information

about the characteristics of the pattern that represents.

Such recognition is diffused through all the dendrites

of which the recognition module is connected. When

it is not recognized, it sends a signal that maybe

involves learning.

According to the hierarchical architecture of

Ar2P, the hierarchy of patterns to recognize the style

of driving, would be as follows: at the first level X1

are the pattern recognition modules of emotions of the

Different Intelligent Approaches for Modeling the Style of Car Driving

287

driver, at the second level X2 are the pattern

recognition modules of drive state, and finally, at the

level X3 the pattern recognition modules of style of

driving. Table 7 shows the mathematical formulation

of the recognition model. Ar2P uses dynamic pattern

recognition modules, what contains the information

needed to recognize a pattern (descriptors, weight,

etc.). Table 8 represents the structure of a pattern

recognition module (Puerto & Aguilar 2017).

Table 7: Mathematical Formulation of the Recognition

Problem.

# Equation

A dynamic pattern is formally defined as a 3-tuple

is a vector that collects all the n descriptors of d: d

denotes a characteristics descriptor, and d

denotes a

perception descriptor.

It is the domain vector of each characteristics descriptor

It is the domain vector of each perception descriptor

They are the change functions, specific to each descriptor.

It is a vector of “change event” of each descriptor.

Table 8: Structure of a dynamic pattern recognition module

(Γρ).

S C

Signal State Pointer (P) Weight (W)

1 F Pointer

[0,1]

... … … …

N F Pointer

[0,1]

1 F Pointer

[0,1]

… … … …

M F Pointer

[0,1]

U: <ΔU1, ΔU2>

S: S=<Signal, State> is an array that represents the set

of signals (descriptors) that conform to the pattern

recognized by Γρd and their respective states. The

state variable is "true" when the signal is present and

"false" otherwise. C= <P, W>, P are pointers to the

time series Δτd

(characteristics descriptor) and Δτd

(perception descriptor). The weight column (W)

contains the value of the descriptor importance in the

recognition. U: is the thresholds vector used by the

module (Γρd) to recognize its respective pattern.

There are two types of thresholds: ΔU1 is the

threshold for the recognition by key signals of

characteristics or perception, and ΔU2 is the threshold

for the recognition by partial or total mapping of

signals of characteristics or perception. Each module

produces a recognition signal (So), or petition signal

towards lower levels. So as petition becomes the input

signal s() for the pattern recognition modules of the

lower levels. When there's a recognition signal, it is

distributed to its higher levels attainable.

Suppose we would like to recognize an "aggressive"

pattern. Table 9 shows the instantiation of the first

level of the pattern recognition module in this case.

Table 10 shows the instantiation of the last level of

the pattern recognition module, for the case where the

driver emotion is "anger". For the rest of the emotions

of the driver, this last level is similarly instantiated.

The instantiation of the second level, for the states of

the driver, is similar.

Table 9: Structure of a dynamic pattern recognition module

for the aggressive pattern: Γρd=aggressive.

S C

Signal State Domain values Weight

Road rural F < Road rural>

0.5

Road urban F < Road urban>

0.6

Stressed F < Stressed>

0.8

Anger F < Anger>

0.8

Rain F <Rain>

0.5

Damage road F < Damage road>

0.6

Does not

follows traffic

signs

F < numerous traffic

tickets, reckless

driving, DWI (driving

while intoxicated),

DUI (driving under the

influence), etc.>

0.8

ICINCO 2017 - 14th International Conference on Informatics in Control, Automation and Robotics

288

Table 10: Structure of a dynamic pattern recognition

module for the driver emotion pattern: Γρd= Anger.

S C

Signal State Domain Weight

The car driver

pulls the door

F < sound of strong door

0.6

High speed F < 150 Km/s ≥ speed ≤

200>

0.8

Strong pressing

the steering-

wheel

F < grade of pressing the

steering-wheel>

0.6

Pupil dilation

high

F < pupil diameter from 6

to 9 mm >

0.6

Heart rate high F < from 200-100

beats/min >

0.8

Mechanical

failure

F < mechanical failure

considered >

0.5

Electrical

failure

F < electrical failure

considered >

0.5

The driver is

shouting

F < The driver is shouting

0.8

The driver is

serious

F < The driver is serious

0.8

Driver moves

violently

F < violent movements

management considered

0.6

U: <ΔU1, ΔU2>

3.3 Based on Fuzzy Logic

A fuzzy controller is a rule-based fuzzy system,

composed of a set of inference rules of the type IF

<Condition> THEN <Action>, that defines the

control actions according to several ranges of the

controlled variables in the problem. Before these

rules can be used, all input signals must be converted

into linguistic/fuzzy variables. In general, the basic

structure of a fuzzy inference system consists of three

conceptual components: a rule base, which contains

the fuzzy rules; a set of fuzzy variables, each one

defined by a set of membership functions; and a

reasoning mechanism that performs the inference

procedure.

We propose to instance the hierarchical

multimodal model of style of driving, using a

Multilayer Fuzzy Classifier System (MFCS). In

(Camargo & Aguilar 2014) is presented a MFCS that

consists of a number of fuzzy systems hierarchically

distributed, which have the advantage that the total

number of rules of the knowledge base is smaller, and

simpler than a conventional fuzzy system. The output

of a Fuzzy Classifier System (FCS) is the input to the

next FCS.

In Figure 1, we show our MFCS model for the

recognition of the style of driving, which is composed

by three FCS, a) a FCS to recognize the emotional

state, b) a FCS to recognize the state of the driver, and

finally, c) a FCS to recognize the style of driving.

The inputs are the same descriptors defined in the

section 2 for each level, but in this case are defined as

fuzzy variables. With these fuzzy variables, we can

describe the set of fuzzy rules of each FCSi.

Figure 1: MFCS Model to recognize styles of driving.

For example, for the case of the FCS1, some of the

possible fuzzy rules are:

• If (use-horn is excessive) and (heart rate is

high) and (facial expression is very serious),

then (driver-emotion is anger).

• If (driver hits steering wheel) and (voice is

high) and (facial expression is serious), then

(driver-emotion is very anger).

For the case of the FCS3, some of the possible fuzzy

rules are:

• If (driver state is very stressed) and (emotion

is anger) then (style-of-driving is aggressive).

• If (driver state is stressed) and (weather is

raining) and (road has potholes) then (style-of-

driving is aggressive).

In the case of the FCS2, the fuzzy rules are similar.

4 COMPARISON OF

APPROACHES

In this section, we perform a qualitative comparison

considering the capabilities of each technique in three

safety-related states (Huang et al. 2010).

4.1 Counter Steering Strategies

(Reasoning Capabilities)

It consists in detecting the negative styles of driving

(aggressive, etc.), in order to guide the driver into a

positive style of driving, for safe driving:

Chronicles: We can observe the process of reasoning

based on temporal logic in a natural way with the

chronicles. For example, a pattern of an emotion like

Different Intelligent Approaches for Modeling the Style of Car Driving

289

the sadness is defined by a set of events at different

times, as facial expressions of type "eyes and

eyebrows with tears that arrive at time T1, and the

voice event " low volume" that occurs at time T2.

That is, the reasoning mechanism is based on the

events of the descriptors and their temporal

relationships, and it manages the incertitude

according to when the events occur.

Ar2P: has the ability to deal with uncertain

knowledge. This is achieved within the structures of

representation of the pattern (i.e., the pattern

recognition modules) using, among other things, the

notion of weight of the descriptors. Particularly, these

modules use meta-variables, such as weights and

value domains, which support different forms or

changes in the descriptors of a pattern. At the level of

the reasoning mechanism, it allows inferring a

situation, and navigating among the modules.

Fuzzy Logic: allows an approximate reasoning,

which implicitly can manage the incertitude, using

the idea of imprecision and information granularity in

the definition of the fuzzy descriptors of our

multimodal pattern model. The fuzzy theory provides

a mechanism for representing linguistic constructs,

such as “many,” “low,” “medium,” “often,” “few”.

Fuzzy logic provides an inference structure that

enables the utilization of these constructs in our fuzzy

descriptors, through our MFCS. Additionally, our

MFCS is an excellent strategy to describe the

different levels of our pattern model. Finally, it can

convert linguistic strategy into control actions, based

on the diagnostic process inferred.

4.2 Adaptation Strategies (Learning

Capability)

It consists in the capability of a quick adaptation to

the personality of the driver:

Chronicles: A same situation (an emotion, a style of

driving, etc.) can be described by different chronicles,

to express the diversities of contexts where a same

situation can occur (for example, an aggressive

behavior). However, the main problem is to learn the

set of chronicles required. In the literature, there are

two types of learning process in the chronicles

paradigm (Aguilar 2011): to learn the structure of a

chronicle, or to parameterize a general chronicle. This

is an open problem. In a real system like our

proposition, we can define general chronicles for each

descriptor of our model, and then parameterize these

chronicles to each driver. This approach requires a

robust chronicle database, which would be constantly

learned to adapt them to the driver and new situations.

Ar2P: uses two strategies of adaptation (Puerto &

Aguilar 2016) the first one, called new learning,

occurs when the input pattern was not recognized

(there is not a module that recognizes it). The second

one, called reinforcement learning, occurs when the

input pattern was recognized. These two learning

mechanisms allow a quick adaptation to the style of

driving of the driver. On the other hand, AR2P

paradigm has the ability to adapt their pattern

recognition modules in accordance with the

recognized patterns, readjusting the importance of the

weights, in order to improve the management of the

incertitude.

Fuzzy Logic: A FCS can learn the rules and the

structures of the fuzzy variables. That means, the

membership functions of the fuzzy variables can be

adapted to the context, and the rules of the database

can be modified (their antecedent and consequent

components) (Camargo & Aguilar 2014). For

example, when the fuzzy definition of the happy

emotions is not adequate, the membership functions

can be modified. Similarly, for the case of the fuzzy

rules, the rules must adapt to reflect the specific

patterns of each individual (maybe, the reasons of an

aggressive behavior of an individual can be very

different with respect to other individuals). To

achieve this, the FCS allows the modification of the

rules when it is presented new information.

4.3 Communication of the Driver’s

Emotional State (Communication

Capabilities)

In this case, we like to evaluate the scenario of the

IoT, where the exchange of information is natural

between heterogeneous devices, such as two vehicles.

Chronicles: The communication within different

chronicles are events. That is valid for the case where

the chronicles are in the same car, or in different cars.

These events can include specific information

required by the chronicles, but it is the only

information required. The hierarchical model of

driving patterns communicates the required messages

with the events generated by the different descriptors,

or chronicles recognized, which contains relevant

information about how drives a driver, in order to

generate the actions concerning with the recognized

driving style. In the previous scenario, the vehicle 1

(v1) sends an event to inform that the driver is “falling

asleep”, to the rest of the vehicles.

Ar2P: only need to send the signals about the

recognition of a given descriptor (for example, the

emotional state of the driver). This signal is the input

of one of the modules of recognition in the other

ICINCO 2017 - 14th International Conference on Informatics in Control, Automation and Robotics

290

vehicle, such as the complex pattern is not important

to know. In the previous scenario, the vehicle 1 (v1)

sends a signal to inform that has recognized the driver

is “falling asleep” to the rest of the vehicles.

Fuzzy Logic: In this case, we have two

possibilities: to send a discrete value, which must be

defuzzifiered in the other vehicle (that is, the output

fuzzy descriptor must be defuzzifiered and sent to the

other vehicles), to send the values of the fuzzy

variables (but on the other side the fuzzy system must

be similar). The main problem is that we can have

multiple outputs (multiple active rules, which can

represent several styles of driving active), and they

must be sent to the other vehicles in order to have a

real idea of the context.

5 CONCLUSIONS

In this paper, we have proposed a hierarchical pattern

of the style of driving, which consider 3 levels of

recognition, one to recognize the emotional state,

other to recognize the state of the driver, and finally,

the last one corresponds to the style of driving. Our

model is flexible because it allows incorporate new

descriptors in the model, for example, about the

traffic flow, among other things.

In addition, the paper analyses three techniques to

recognize the style of driving, one based on fuzzy

logic, another based on chronicles, and other based on

Ar2P. We have compared these techniques in 3 cases:

for defining countersteering strategies, or its adaptive

capability to the driver, or to communicate the style

of driving of the driver recognized. Each technique

has its advantage and disadvantage, and depend on

the real context (IoT) to choose to one of them.

As future work, we will carry out the

implementation of these techniques in a simulated

context, to measure the three previous criteria using

specific metrics for each one. In this way, we will

carry out a quantitative comparison, which is

complementary to the qualitative comparison

analysed in this work.

ACKNOWLEDGEMENTS

Dr Aguilar has been partially supported by the

Prometeo Project of the Ministry of Higher

Education, Science, Technology and Innovation of

the Republic of Ecuador.

REFERENCES

Aguilar, J., 2011. Temporal Logic from the Chronicles

Paradigm : learning and reasoning problems, and its

applications in Distributed Systems. Ed. Lambert.

Aguilar, J., Chavez, D. & Cordero, J., 2016. A Dynamic

Recognition Approach of Emotional States for Car

Drivers. In Second International Conference on Techno-

logies and Innovation. Communications in Computer

and Information Science. Springer, pp. 155–168.

Aypar, Y.C. et al., 2014. Clustering of emotional states

under different task difficulty levels for the robot-

assisted rehabilitation system-RehabRoby. In 11th

International Conference on Informatics in Control,

Automation and Robotics (ICINCO), pp. 34–41.

Camargo, E. & Aguilar, J., 2014. Hybrid intelligent

supervision model of oil wells. In IEEE International

Conference on Fuzzy Systems. pp. 934–939.

Cordero, J. & Aguilar, J., 2016. Reconocimiento multimo-

dal de emociones en un entorno inteligente basado en

crónicas. Gráficas El Portatítulo, pp.525–541.

Eyben, F. et al., 2010. Emotion on the road—necessity,

acceptance, and feasibility of affective computing in the

car. Advances in human-computer interaction.

Guoying, C. & Danpan, W., 2016. Study on Identification

of Driver Steering Behavior Characteristics Based on

Pattern Recognition. International Robotics and

Automation Journal, 1(1), pp.1–8.

Huang, C.-L. et al., 2010. Adaptive intervehicle

communication control for cooperative safety systems.

IEEE network, 24(1).

Katsis, C.D. et al., 2015. Emotion recognition in car

industry. Emotion Recognition: A Pattern Analysis

Approach, pp.515–544.

Kolli, A. et al., 2011. Non-intrusive car driver’s emotion

recognition using thermal camera. In Joint 3rd Int’l

Workshop on Nonlinear Dynamics and Synchronization

(INDS) & 16th Int’l Symposium on Theoretical

Electrical Engineering, pp. 1–5.

Lin, N. et al., 2014. An overview on study of identification

of driver behavior characteristics for automotive

control. Mathematical Problems in Engineering.

Paschero, M. et al., 2012. A real time classifier for emotion

and stress recognition in a vehicle driver. In IEEE

International Symposium on Industrial Electronics, pp.

1690–1695.

Puerto, E. & Aguilar, J., 2016. Learning Algorithm for the

Recursive Pattern Recognition Model. Applied

Artificial Intelligence, 30(7), pp.662–678.

Puerto, E. & Aguilar, J., 2017. Un algoritmo recursivo de

reconocimiento de patrones, Revista Técnica de

Ingeniería de la Universidad del Zulia, 40 (2).

Tawari, A. & Trivedi, M., 2010. Speech based emotion

classification framework for driver assistance system.

In IEEE Intelligent Vehicles Symposium), pp. 174–178.

Wang, J. et al., 2013. An adaptive longitudinal driving

assistance system based on driver characteristics. IEEE

Transactions on Intelligent Transportation Systems,

14(1), pp.1–12.

Different Intelligent Approaches for Modeling the Style of Car Driving

291