Effect analysis of emotions on driving intention in two-lane environment

Abstract

Driving emotion plays an important role in intention generation and behavior decision. It is an important premise to understand the influencing mechanism of driving emotion on intention correctly for the realization of driving intention identification, active vehicle security warning, and intelligent driving. In the case of two-lane, multi-source dynamic data of human–vehicle–environment under drivers’ different emotional states is obtained through actual and virtual driving experiment in this article. Considering the influence of emotion on intention, the probability distributions of the driving intention in different emotions are obtained by means of the immune algorithm. And the priming effect of vehicle driving emotion on intention was analyzed deeply. The research results can provide theoretical foundation for the research of driving intention identification, active vehicle security warning system, and intelligent driving command system under the condition of Internet of things.

Keywords

Driving emotion driving intention effect analysis immune algorithm active vehicle safety warning system

Introduction

With the rapid development of economy in China, vehicle population, especially the number of private cars, is increasing rapidly. In the road traffic system, the contradiction of human–vehicle–environment is becoming more and more serious,¹ and traffic accidents happen frequently. According to expert analysis results, the influence of human factors is as high as 90% in the traffic accident, among which 70% are caused by drivers.² Controlling driver behavior is still an active factor in improving security of automobile.³ Emotion is the psychological feedback in the process of anticipation and judgment of the related things in the environment, which influences the physiological behavior of human being.⁴ Existing research indicates that driving emotion will affect driving behavior in driving process, and driving behavior is driven by driving intention. The influence of emotion on driving intention cannot be ignored in driving behavior research.⁵ It is of great significance to reveal the mechanism of emotion on intention accurately for the research of driving intention identification, active vehicle security warning system, and intelligent driving command system under the condition of Internet of things. Through analysis result of literatures, at home and abroad, the research of emotion–intention interaction is mainly focused on sports,^6–9 shopping consumption,^10–15 environmental protection,^14,16–18 moral education,^19–21 and so on. For example, emotional factors had been added to traditional sports intention identification models by C Mohiyeddini et al.,⁶ and the results showed that it can improve the recognition ability of traditional sports intention identification model, if emotional factors were considered. After controlling for the theory of planned behavior constructs, X Wang⁷ analyzed the moderating role of past behavior on the relationship between anticipated negative emotions and intentions. The results showed that the inclusion of past behavior can help better delineate the conditions under which anticipated emotions predict behavioral intentions. SMC Loureiro and Roschk¹⁰ proposed a model to examine the effect of the atmospheric cues graphics design and information design on positive emotions and loyalty intentions. The results showed that in the offline context positive emotions predict loyalty among younger customers but not among older ones; in the online context the effect of graphics design on loyalty is stronger. The structural equation model had been used by HB Chen et al.¹¹ to study the influence of nostalgia on restaurant consumer intention. The results showed that nostalgia can not only promote the consumption intention but also weaken the consumption intention. According to the Mehrabian Russell Model, SH Tsaur et al.¹² studied the influence of aesthetic labor performance on the positive emotions and behavioral intentions of customers. The results showed that aesthetic labor performance had positive influence on customer’s positive emotion and behavior intention, and positive emotion had guiding function on customer behavior intention. J Wang and Wu¹³ used the partial least squares structural equation model to study the relationship between consumer emotions and consumer intention. The emotions include pride, guilt, respect, and anger, and the purchases include energy-saving appliances and non-energy saving appliances. The results showed that pride, guilt, and respect for the three emotions had a positive effect on the two consumer intentions, while anger emotion only had a greater impact on the consumer intention of non-energy saving appliances. R Ladhari et al.¹⁴ studied customer purchase intentions through customer perceptions of service quality and emotional responses to receiving services. The result showed that both service quality and service environment can affect customer satisfaction, and the higher the emotional satisfaction is, the stronger the purchase intention is. R Ladhari et al.¹⁴ researched the role of emotions in utilitarian service settings. The results showed that perceived service quality and service environment both increase positive emotional satisfaction, and positive emotional satisfaction leads to a high perception of product quality, high recommendation, patronage intention, and likelihood of purchase. NS Harth et al.¹⁶ compared and analyzed the predictive effects of three group emotions, including pride, guilt, and anger, on three intentions related to the natural environment. The result showed that guilt feeling makes people more inclined to have the intention to repair the environment, anger feeling makes people more inclined to have the intention to punish the environmental destroyer, and pride feeling makes people more inclined to have the intention to protect the environment. J Yang et al.¹⁷ discussed the relationship between the festival visitors’ behavior intention, consumption emotion, and perceived value. And they also analyzed the moderating effect of festival environmental on the influence of tourists’ consuming emotion and perceived value on behavior intention. S Chen²¹ used the clinical interview method of story situation to investigate the influence of different situations on the judgment and attribution of moral emotion in children aged 4 to 6. In the field of traffic, the study of driver’s intention by experts at home and abroad is mainly to identify driver’s intention by using road curvature, lane line position, and other factors. However, the interaction between driver’s emotion and intention was not found. For example, Dogan et al.²² explored the effect of machine learning on driver’s intention identification. The variables of road curvature, lane position, steering wheel angle, lateral acceleration, collision time, and so on were combined, and the artificial neural network model and support vector machine were used to predict the driver’s behavior. In the research of J Peng et al.,²³ a prediction index system for left lane change was constructed by considering drivers’ visual search behaviors, vehicle operation behaviors, vehicle motion states, and driving conditions. And a back-propagation neural network model was developed to predict lane-changing behavior. In the research by L Ohashi et al.,²⁴ the data such as driver’s motions, speed of the vehicle, and the distance between the vehicle and the intersection were analyzed, and the concept fuzzy set theory was used to identify driver’s intention (e.g. turn left, turn right, or go straight).

The movement of the vehicle is determined by driving behavior, which keeps the driver’s life safety.²⁵ C Ma et al.²⁶ explored the influence of driving behavior on driver’s life safety. Diving intention changes with the real-time change of emotion and environment, and it guides the driving behavior. Human cognition and behavior are regulated by emotion,⁴ and all purposeful behaviors are carried out under the influence of emotion. It is helpful for improving road traffic safety to study the influence of emotion on intention.^27,28 However, many scholars of domestic and overseas often neglect the influence of emotion on intention in the aspects of studying driving intention, which cause difficulties in accurately reflecting the true mental state of drivers and intention identification. In view of this, the complexity of vehicle marshaling is analyzed in this article in two-lane environment. And multi-source dynamic data of human–vehicle–environment under drivers’ different emotional states is obtained through actual and virtual driving experiment. Effect analysis model of emotions on drivers’ intention based on immune algorithm is established, and the influence of different emotions on driving intention is deeply analyzed.

Method

Analysis of the complexity of vehicle marshaling

The vehicle marshaling relationship is composed of dynamic traffic entities, and it has important influence on driver’s behavior decision.^29,30 The research scenario is defined to analyze the impact of driver emotion on intention in different vehicle marshaling relationships (as shown in Figure 1) better. When the target vehicle is running in the right lane, the vehicles in the driver’s interest-sensitive area (defined as the area with greatest impact on vehicle safety and the driver’s attention) are divided into target vehicle u (a car that is driven by a test is called a target vehicle), forward vehicle $u_{1}$ , left front vehicle $u_{2}$ , left rear vehicle $u_{3}$ , and rear vehicle $u_{4}$ . The standard of vehicle location is the position of the front bumper $(l)$ and the line between the driveways of the same direction (the front and rear directions are divided by the l, and the left and right directions are divided by lane line).³¹ When the target vehicle is running in the left lane, the vehicles in the drivers’ interest-sensitive area are divided into target vehicle u, forward vehicle $u_{5}$ , right front vehicle $u_{6}$ , right rear vehicle $u_{7}$ , and rear vehicle $u_{8}$ . In the processing of driving, driver would choose safer driving behavior according to the distribution of the interfering vehicles in drivers’ interest-sensitive area, driving state, and self-condition, that is, the interfering vehicle in interest-sensitive area will attract or repel the target vehicle to drive in the area. With the introduction of vehicle marshaling, the process of analysis would be simplified and the results would be more accurate.

Figure 1.

Position of vehicle around target vehicle in two-lane environment: (a) the target car is in the right lane; (b) the target car is in the left lane.

The 32 vehicle marshaling relationships are constructed in two lanes, based on the above scene definition method. The interfering vehicles at different locations have different effects on the target vehicle drivers.^32,33 When the target vehicle is moving in a two-lane environment, the vehicle marshaling relationship is shown in Figures 2 and 3.

Figure 2.

Vehicle marshaling relationship in two-lane environment when the target vehicle runs in the right lane.

Figure 3.

Vehicle marshaling relationship in two-lane environment when the target vehicle runs in the left lane.

Effect analysis model of emotions on driving intention based on immune algorithm

Introduction of immune algorithm

The biological immune system is simulated by the immune algorithm,³⁴ and it is an algorithm with some characteristics, such as memory mechanism, regulation mechanism, and evaluation mechanism and produces some kind of specific antibody intensively. This algorithm is implemented by multi-point parallel computation. Excellent solutions are memorized and retained, and local convergence problems are effectively avoided.³⁵ The functions of immune algorithm such as immune memory, self-regulation, and diversity preservation are exploited by immune-genetic algorithm based on information entropy. The affinity and concentration of antibodies are expressed by information entropy. The diversity of the real immune system is reflected. And the self-regulation function of immune response is manifested by promoting or inhibiting the production of antibodies.^36–38 Probability distribution of the driving intention given the emotion status is obtained by immune-genetic algorithm, and, based on that, effect analysis model of emotions on driving intention is established. The algorithm flow is shown in Figure 4.

Figure 4.

Flowchart of immune algorithm based on information entropy.

Modeling

The immune-genetic algorithm combines the advantages of biological immune principle and genetic algorithm. In this study, seven kinds of driving intention such as keep the speed straight (It1), speed up straight (It2), speed down straight (It3), keep the speed lane change (It4), speed up lane change (It5), speed down lane change (It6), and stop (It7) are regarded as antigens, and, correspondingly, eight kinds of common emotions such as anger (Em1), surprise (Em2), fear (Em3), anxiety (Em4), helpless (Em5), flout (Em6), relaxation (Em7), and joviality (Em8) are regarded as antibodies. The affinity and concentration of antibodies are represented by information entropy. The driving emotion set is represented in binary coding. (Character set encoding of antibodies is shown in Figure 5.) It can be supposed that there are n antibodies in the immune-genetic algorithm based on information entropy, and each antibody has m genes. The character set of each bit in genes is ${0, 1}$ , in which 1 represents a certain emotion has been recognized and 0 represents the contrary. Parameter $H_{j} (η)$ is employed to express the information entropy of the j gene and it can be acquired from equation (1)

H_{j} (η) = - \sum_{i = 1}^{S} p_{ij} \log p_{ij}

(1)

In the equation (1), $η$ is compare foundation parameters $(η = 2, 3, \dots, n)$ . $p_{ij}$ is a probability that character i appears on gene locus j in character set. $p_{ij} = α_{ij} / η$ ( $α_{ij}$ is the total number of character i appears on j gene locus, and i = 1,2.) The mean entropy of antibody diversity can be obtained from equation (2)

H (η) = \frac{1}{m} \sum_{j = 1}^{m} H_{j} (η)

(2)

Usually, $η = 2$ . The affinity of antibody u and v is obtained according to the definition of information entropy. As shown in equation (3)

A_{uv} = \frac{1}{1 + H (2)}

(3)

In the equation (3), $H (2)$ is average information entropy of antibody u and v. The range of $A_{uv}$ is (0, 1). The larger $A_{uv}$ is, the higher the similarity between the two antibodies will be. $A_{uv} = 1$ means that the coding genes are identical. The concentration of antibody is shown in equation (4)

d_{i} = \frac{β_{i}}{n}

(4)

In the equation (4), $β_{i}$ is the number of antibodies which affinity is greater than $λ$ compared with antibody i. And $λ$ is the affinity constant in equation (4). Usually, $0.95 ⩽ λ < 1$ .

The corresponding selection mechanism was adopted to promote and inhibit antibody production, depending on the concentration of each antibody. The selection probability $p_{i}$ of antibody i is shown in equation (5)

p_{i} = α p_{fi} + (1 - α) p_{di}

(5)

In equation (5), $α$ is constant adjustment factor, $f_{i}$ is fitness of antibody, and it is the matching degree of antigen and antibody. $p_{fi}$ is the fitness probability of antibody. $p_{fi} = f_{i} / \sum_{i = 1}^{n} f_{i}$ . Based on the concentration of antibody, the inhibition probability of antibody concentration $p_{di}$ can be obtained from $p_{di} = e^{- d_{i}}$ . And the selection probability of antibody i is shown in the equation (6)

p_{i} = α \frac{f_{i}}{\sum_{i = 1}^{n} f_{i}} + (1 - α) \times e^{- d_{i}}

(6)

As is known from equation (6), selection probability is related to both fitness probability of antibody and conditions of antibody concentration. In this way, the diversity of individuals is maintained while antibodies with higher fitness are retained. The constant adjustment factor could change the power of each part of selection probability. It can be seen from this, under certain conditions of antibody concentration, the larger fitness probability of antibody is, the greater selection probability will be; under certain conditions of antibody fitness, the larger antibody concentration is, the smaller selection probability will be. This could effectively prevent some individuals from controlling the selection process, suppress individual premature phenomenon that all individuals in the group are trapped in the same extreme value and stop the evolution, and enhance the local search ability. In this way, the diversity of individuals is maintained while antibodies with higher fitness are retained.

Figure 5.

Character set encoding of antibodies.

According to the above contents, effect analysis model of emotions on driving intention based on immune algorithm is established (shown in Figure 6). Driving intention is comprehensively evaluated by means of analyzing driver characteristics, vehicle and environment, and video replaying of the experiment process, driver questioning, and so on. The fitness of each driving intention under different emotional states is scored in two-lane environment. And the larger the fitness is, the higher the scores will be. The probabilities of drivers’ different intention in an emotional state are obtained with above calculation method.

Figure 6.

Effect analysis model of emotions on driving intention based on immune algorithm.

Experimental design and data processing

Experimental design

The actual driving experiment environment is more close to the natural driving. The data from actual driving experiment can reflect the driver’s real state better. But the experimental is difficult to organize and the cycle is longer. Although there are some differences between virtual driving experiment environment and actual driving experiment, the virtual driving experiment has few restrictions and shorter cycles.

Seven kinds of driving intention such as keep the speed straight (It1), speed up straight (It2), speed down straight (It3), keep the speed lane change (It4), speed up lane change (It5), speed down lane change (It6), and stop (It7) and eight kinds of common emotions such as anger (Em1), surprise (Em2), fear (Em3), anxiety (Em4), helpless (Em5), flout (Em6), relaxation (Em7), and joviality (Em8) are used for this study. The driving intentions are selected by analyzing the vehicle driving status in two-lane environment.^27,28 The driving emotions are selected by referring to negative and positive emotions put forward by Victor S Johnston,^39,40 and the survey of common emotions during driving. The experimental design is shown in Table 1, and some driving emotion visual stimulus material is shown in Figures 8 and 9.

Table 1.

Experimental design.

Items	Set
Items	Actual driving experiment	Virtual driving experiment
Experimental equipments	Road traffic comprehensive experimental vehicle consisted of precision positioning system, velocimeter, lidar, distance laser sensor, human factor engineering experiment system, video-capturing system, etc. and other 4 cars (Vehicles that interferes with the running of the target vehicle. And they are used to form different vehicle marshaling states)	Human factor engineering experiment system, interactive parallel driving virtual experiment platform, video-capturing system.
Route	The part of the Zhangzhou road between Jiangmeng road and Shanshen line in Zhangdian District of Zibo city Shandong province (shown in Figures 7)
Experimental materials	Psychometric questionnaire of driver’s propensity,^24,25 Psychometric questionnaire of willpower,^36,37 International Affective Picture System (IAPS), Chinese Affective Picture System (CAPS)
Time	8:00~18:00 of normal working day, and the weather is fine.
Subjects	Fifty-four persons are selected as the experimental subjects, including 27 males and 27 females. Their age range is between 18 and 70 years old. The average age is 33.5 years old. The average driving mileage is 12,000 kilometers
Methods of emotion stimulation	Different emotions of the driver are stimulated by pictures, video, audio, and other materials in IAPS and CAPS (some of the emotional materials are shown in Figures 8 and 9). The driving emotions such as fear, helplessness, anxiety, and anger are stimulated by some traffic scenes, such as setting the traffic accident, traffic congestion, wait for a long time in intersection, crowded vehicle road, and the target vehicle is disturbed by interfering vehicles in interactive parallel driving virtual experiment platform. The driver’s flout emotion is stimulated using the driving simulator to build the driving race scenes. Some driving emotions such as relaxation and joviality are stimulated by communicating with the subject, giving some gifts, or smiling, singing, etc.
Methods of emotion prolonging	The driving emotions are prolonged by listening to the music about the same emotions, communicating, memory, etc. during the vehicle’s driving process

Figure 7.

Actual driving experiment route.

Figure 8.

Partial driving emotion visual stimulus material.

Figure 9.

Part picture of joviality emotion face.

Tests of driver physiology and psychology

Basic information of drivers such as age and gender are counted. The drivers’ reaction times are obtained by physiological test and the test method used by Yang.⁴¹ Drivers’ propensity (conservative, common, radical) and willpower (strong, middle, weak) types are determined by weighted calculation, based on driving propensity questionnaire and willpower test questionnaire.

Driving experiment

Drivers are familiar with the experimental vehicle (environment including the actual and virtual experiment vehicle) before the driving test begin. Drivers listen to soothing music before emotional stimulation. When drivers are in a relaxed state, their emotions are stimulated through pictures, video, audio, and other materials in IAPS and CAPS. Driving emotions are determined according to their facial expressions and inquiry results. And driving experiments start after determining the driver’s feelings. According to different emotional characteristics, in addition to using the above material to stimulate emotions, a variety of methods such as spiritual motivation and material reward are also used (shown in Table 1). The vehicle environment dynamic data in different vehicle marshaling is obtained synchronously by instruments. In the running process of vehicles, the Jamming vehicles are used to form different vehicle marshaling relationships. And the driver’s facial expressions and movements are videotaped. The driving experiment scene is shown in Figures 10 and 11.

Figure 10.

Actual driving experiment.

Figure 11.

Virtual driving experiment.

To ensure the effectiveness of the induction method and the scientific of the experimental design, the experiment simulations had been done many times. In the experiment simulations, the experimental organizers were seen as experimental objects by turns. The 54 selected objects drives mentioned in this article were not used in the experiment simulations, because some emotion induction method could not use many times. For example, it is easy to induct one’s joviality emotion using material reward method at the first time; however, the driver may take the prize for granted when the method was used over and over and the joviality emotion would not be inducted. In the experiment simulations, the emotion induction method for each emotion and every aspect of induction experiment was adjusted and revised constantly until the optimal effect is achieved.

The result of emotion induction affected the validity and reliability of the research directly. The natural speech questioning, emotional self-report, emotional survey scale, analysis of facial expressions, and behavioral actions during driving were utilized comprehensively to assess the result of emotion induction in this article. The driver’s emotions were mainly labeled by emotional survey scale, and the natural speech questioning, emotional self-report, analysis of facial expressions, and behavioral actions during driving were used to verify the validity of emotional survey scale. If inconsistent results of emotion induction through different valuation methods were got, the experiment data would be seen as invalid data and would be removed. The CAPS adjective lexicon contains 500 adjective words, and the pleasure, arousal, and dominance degree index of each word was assigned by the 9-point scale. All the words were divided into five grades according to the pleasure, arousal, and the dominance degree that correspond to the five grades of the pleasure, arousal, and dominance degree of PAD emotion model. Three words were selected separately from different grades words of pleasure, arousal, and dominance degree (totally five groups) to form the grades survey scale of pleasure, arousal, and dominance degree. Experimental objects’ facial expressions and behavior during driving were recorded in real time with video detection system. The experimental objects’ emotion experience was inquired through natural speech questioning, and the results of the inquiry were recorded. After the driving experiment, the experimental objects would watch the video playback and report their emotional experience at different moments of the video. The experimental objects also needed to select the word groups that were most close to actual emotional experience at different moments from the grades questionnaires of pleasure, arousal, and dominance degree according to the meaning of words. The effectiveness of emotion inductions was determined finally according to the results of natural speech questioning, emotional self-report, emotional survey scale, the analysis of facial expressions, and behavioral actions during driving through the above methods.

Data processing

According to the above method, drivers’ emotions are stimulated, and then the drivers are organized to conduct driving experiments. After the driving test, the drivers watch the driving video immediately and report the emotional experience of each time period to the experimental organizer. Combined with emotion self-report, natural speech questioning, emotional survey scale, analysis of facial expression, and behavioral action during driving were utilized comprehensively to assess the effectiveness of emotion induction. After analyzing a large amount of data, each emotion is divided into three kinds: strong, middle and weak. Based on emotional screening, the emotion–intention combination units of these eight emotions and seven driving intentions are counted in different vehicle groups. And each data unit corresponds to an emotion–intention combination and the corresponding data of driver characteristics, vehicle, and driving environment. In real vehicle and virtual driving experiments, 9384 and 77,261 affective emotion–intention combination units are obtained under 32 vehicle marshaling relationships. And these data together form the original database; part variables and experimental data are shown in Tables 2 and 3.

Table 2.

Partial variables and symbols.

Variable	Symbol	Variable	Symbol	Variable	Symbol
Gender	G	Drivers’ reaction times $(ms)$	$O_{t}$	Target vehicle acceleration $(m / s^{2})$	a
Age $(year)$	A	Acceleration frequency of target vehicle $(sub / \min)$	$f_{a}$	Acceleration interference of target vehicle $(m / s^{2})$	$σ$
Driving mileage $(km)$	$D_{A}$	Deceleration frequency of target vehicle $(sub / \min)$	$f_{b}$	vehicle marshaling relationship	T
Emotional state	Em	Lane changing frequency of target vehicle $(sub / \min)$	$f_{c}$	Distance between target vehicle and road centerline $(m)$	$d_{1}$
intention	It	Average depth of accelerator pedal	$T_{a}$	Speed of jamming vehicle i $(km / h)$	$v_{i}$
willpower	E	Average depth of brake pedal	$T_{b}$	Acceleration of interference vehicle i $(m / s^{2})$	$a_{i}$
Drivers’ propensity	$D_{s}$	Target vehicle speed $(km / h)$	v	The relative distance between the target vehicle and the disturbance vehicle I $(m)$	$d_{c - i}$
Desired speed $(km / h)$	$v_{e}$	Average rotation angle of steering wheel (°)	w	The gap acceptance	$Δ d$
Expected distance between target vehicle and interfering vehicle $i (m)$	$d_{e - i}$	Average steering power of steering wheel $(N)$	$f_{m}$

Interfering vehicles are vehicles in which the target vehicle is affected directly in interest-sensitive area. i is the number of the interfering vehicle (shown in Figure 1), $i = 1, 2, \dots, 8$ . Acceleration disturbance is a description of vehicle speed swing, a quantitative evaluation index of ride comfort, and its calculation method refer to reference.⁴²

Table 3.

Rules of supplement for traffic participant representative.

Number	G	A	$D_{A}$	Em	It	E	$D_{s}$	$v_{e}$	$O_{t}$	$f_{a}$	$f_{b}$
1	male	26	81520	joviality	keep the speedstraight	strong	conservative	40	1268	13	5
	$f_{c}$	$T_{a}$	$T_{b}$	w	$f_{m}$	v	a	$σ$	$d_{1}$	$d_{c - 1}$	$d_{c - 2}$
	3	0.24	0.089	35.9	20.8	35.63	0.31	2.31	4.52	13.3	15
	$d_{c - 3}$	$d_{c - 4}$	$v_{1}$	$v_{2}$	$v_{3}$	$v_{4}$	$a_{1}$	$a_{2}$	$a_{3}$	$a_{4}$	$d_{e - 1}$
	5	16	40.21	36.54	35.44	32.11	0.10	0.11	0	−0.13	13
	$d_{e - 2}$	$d_{e - 3}$	$d_{e - 4}$	T	$Δ d$	$v_{e} - v$	$d_{e - 1} - d_{c - 1}$	$d_{e - 2} - d_{c - 2}$	$d_{e - 3} - d_{c - 3}$	$d_{e - 4} - d_{c - 4}$
	8	6	8	$T_{16}$	5	4.37	−0.3	−7	1	−8
2	female	28	61165	fear	speed downlane change	middle	conservative	30	1151	8	9
	$f_{c}$	$T_{a}$	$T_{b}$	w	$f_{m}$	v	a	$σ$	$d_{1}$	$d_{c - 1}$	$d_{c - 2}$
	2	0.48	0.158	29.1	17.6	36.74	−1.58	1.59	5.13	15.2	16.3
	$d_{c - 3}$	$d_{c - 4}$	$v_{1}$	$v_{2}$	$v_{3}$	$v_{4}$	$a_{1}$	$a_{2}$	$a_{3}$	$a_{4}$	$d_{e - 1}$
	7.1	10.2	40.32	36.43	32.14	35.11	1.13	1.32	0.12	−1.25	18
	$d_{e - 2}$	$d_{e - 3}$	$d_{e - 4}$	T	$Δ d$	$v_{e} - v$	$d_{e - 1} - d_{c - 1}$	$d_{e - 2} - d_{c - 2}$	$d_{e - 3} - d_{c - 3}$	$d_{e - 4} - d_{c - 4}$
	20	10	20	$T_{16}$	3	−6.74	2.8	3.7	2.9	9.8
…	…	…	…	…	…	…	…	…	…	…	…
n	female	50	893570	relaxation	keep the speed straight	strong	conservative	35	1110	12	6
	$f_{c}$	$T_{a}$	$T_{b}$	w	$f_{m}$	v	a	$σ$	$d_{1}$	$d_{c - 1}$	$d_{c - 2}$
	4	0.59	0. 161	−35.1	20.1	31.05	1.33	1.67	4.33	15.6	18.7
	$d_{c - 3}$	$d_{c - 4}$	$v_{1}$	$v_{2}$	$v_{3}$	$v_{4}$	$a_{1}$	$a_{2}$	$a_{3}$	$a_{4}$	$d_{e - 1}$
	8.2	15.6	38.31	36.43	32.65	35.24	2.05	1.05	0.15	0.21	15
	$d_{e - 2}$	$d_{e - 3}$	$d_{e - 4}$	T	$Δ d$	$v_{e} - v$	$d_{e - 1} - d_{c - 1}$	$d_{e - 2} - d_{c - 2}$	$d_{e - 3} - d_{c - 3}$	$d_{e - 4} - d_{c - 4}$
	20	10	20	$T_{16}$	6	3.95	−0.6	1.3	1.8	4.4

Three experimental objects’ data, which was got from the emotion-inducted driving experiments, are enumerated in Table 3. The data included some typical and relatively stable characteristic parameters of the driver and the movement characteristic parameters of vehicle which was driven by the driver with a kind of emotion. The drivers’ characteristic parameters could represent his or her personal features associated with emotion, such as emotional intelligence, affective sensitivity, and so on. For example, it is generally more difficult for the driver with a relative more years’ driving experience to evoke an emotion in the driving process, because more abundant driving experience made him or her more deliberate to face different kinds of traffic situations. It can be seen from the emotion induction experiments clearly. However, once emotion was evoked, especially negative emotion, the behavioral expression of the driver with more driving experience could be more radical. On the one hand, he or she who was confident in his or her driving skills would pursue a high driving speed. On the other hand, when the speed or space requirements were blocked by other inexperienced drivers, his or her emotion state would become more terrible and risky driving behavior would be taken. The movement characteristic parameters of vehicle were important data to estimate the driver’s physical and psychological state. In the emotion-inducted driving experiment, the differences of driver’s facial expression, speaking, body movement, and so on were obvious in different driving emotion states. For example, a driver with anger emotion generally had a fierce brake and a high driving speed. On the contrary, the driver would be gentler to brake and the driving speed always at a very low level which was slower than his or her desired speed under the state of fear emotion. As can be seen from the experiment data listed in Table 3, the speed values had a big span for one driver in certain emotion state compared with the desired speed. For the NO.1 driver who had a desired speed of 40 km/h, the driving speed was 35.63 km/h under the joviality emotion state. Similarly, for the NO.n driver who had an desired speed of 35 km/h, the driving speed was 31.05 km/h under the relaxation emotion state, whereas for the driver of NO.2, the driving speed under fear emotion states was higher than her expected speed. It may be given rise to the other personal characteristics or experimental environment. The other parameters showed the similar pattern. It should be pointed that the parameters selected in this study were all non-intrusive data of which acquisition process would not affect driving task. It is of great significance for the realization of the real-time and online identification of drivers’ emotions.

In the above variables, G, $T_{d}$ , It, E, and Em are attribute variables, and other variables are quantitative variables. In this article, attribute variable encoding is treated as dummy variable, and quantity variable coding is treated as discrete variable. The settings for each variable are shown in Table 4.

Table 4.

Relevant variable settings.

Symbol	Attribute	Evaluation	Symbol	Attribute	Evaluation	Symbol	Attribute	Evaluation	Symbol	Attribute	Evaluation
$G 1$	male	1	$A_{1}$	$20 ~ 35$	1	$D_{A}^{1}$	$< 219000$	1	$D_{S}^{1}$	conservative	1
			$A_{2}$	$35 ~ 50$	2	$D_{A}^{2}$	$219000 ~ 438000$	2	$D_{S}^{2}$	common	2
$G 2$	female	2	$A_{3}$	$50 ~ 65$	3	$D_{A}^{3}$	$> 438000$	3	$D_{S}^{3}$	radical	3
$E_{1}$	strong	1	$I t^{1}$	keep thespeed straight	1	$I t^{4}$	keep the speedlane change	4	$I t^{6}$	speed down lane change	6
$E 2$	middle	2	$I t^{2}$	speed upstraight	2
$E 3$	weak	3	$I t^{3}$	speed down straight	3	$I t^{5}$	speed uplane change	5	$I t^{7}$	stop	7
$v_{e}^{1}$	$< 40$	1	$d_{e - 1}^{1}$	$< 12$	1	$d_{e - 2}^{1}$	$< 10$	1	$d_{e - 3}^{1}$	$< 5$	1
$v_{e}^{2}$	$40 ~ 50$	2	$d_{e - 1}^{2}$	$12 ~ 18$	2	$d_{e - 2}^{2}$	$10 ~ 20$	2	$d_{e - 3}^{2}$	$5 ~ 15$	2
$v_{e}^{3}$	$> 50$	3	$d_{e - 1}^{3}$	$> 18$	3	$d_{e - 2}^{3}$	$> 20$	3	$d_{e - 3}^{3}$	$> 15$	3
$d_{e - 4}^{1}$	$< 12$	1	$d_{e - 5}^{1}$	$< 12$	1	$d_{e - 6}^{1}$	$< 10$	1	$d_{e - 7}^{1}$	$< 5$	1
$d_{e - 4}^{2}$	$12 ~ 18$	2	$d_{e - 5}^{2}$	$12 ~ 18$	2	$d_{e - 6}^{2}$	$10 ~ 20$	2	$d_{e - 7}^{2}$	$5 ~ 15$	2
$d_{e - 4}^{3}$	$> 18$	3	$d_{e - 5}^{3}$	$> 18$	3	$d_{e - 6}^{3}$	$> 20$	3	$d_{e - 7}^{3}$	$> 15$	3
$d_{e - 8}^{1}$	$< 12$	1	$O_{t}^{1}$	$< 1.05$	1	$f_{a}^{1}$	$0 ~ 5$	1	$f_{b}^{1}$	$0 ~ 5$	1
$d_{e - 8}^{2}$	$12 ~ 18$	2	$O_{t}^{2}$	$1.05 ~ 1.15$	2	$f_{a}^{2}$	$5 ~ 10$	2	$f_{b}^{2}$	$5 ~ 10$	2
$d_{e - 8}^{3}$	$> 18$	3	$O_{t}^{3}$	$> 1.15$	3	$f_{a}^{3}$	$10 ~ 15$	3	$f_{b}^{3}$	$10 ~ 15$	3
$f_{d}^{1}$	$0 ~ 5$	1	$T_{a}^{1}$	$0 ~ 0.33$	1	$T_{b}^{1}$	$0 ~ 0.33$	1	$v_{1}$	$< 40$	1
$f_{d}^{2}$	$5 ~ 10$	2	$T_{a}^{2}$	$0.33 ~ 0.66$	2	$T_{b}^{2}$	$0.33 ~ 0.66$	2	$v_{2}$	$40 ~ 50$	2
$f_{d}^{3}$	$10 ~ 15$	3	$T_{a}^{3}$	$0.66 ~ 1$	3	$T_{b}^{3}$	$0.66 ~ 1$	3	$v_{3}$	$> 50$	3
$w^{1}$	$< 20$	1	$f_{m}^{1}$	$< 10$	1	$a^{1}$	$< - 1$	1	$σ_{1}$	$0 ~ 1.5$	1
$w^{2}$	$20 ~ 40$	2	$f_{m}^{2}$	$10 ~ 20$	2	$a^{2}$	$- 1 ~ 1$	2	$σ 2$	$1.5 ~ 3$	2
$w^{3}$	$> 60$	3	$f_{m}^{3}$	$> 20$	3	$a^{3}$	$> 1$	3	$σ 3$	$3 ~ 4.5$	3
$d_{1}^{1}$	$< 1$	1	$v_{i}^{1}$	$< 40$	1	$a_{i}^{1}$	$< - 1$	1	$d_{c - 1}^{1}$	$< 12$	1
$d_{1}^{2}$	$1 ~ 3.5$	2	$v_{i}^{2}$	$40 ~ 50$	2	$a_{i}^{2}$	$- 1 ~ 1$	2	$d_{c - 1}^{2}$	$12 ~ 18$	2
$d_{1}^{3}$	$> 3.5$	3	$v_{i}^{3}$	$> 50$	3	$a_{i}^{3}$	$> 1$	3	$d_{c - 1}^{3}$	$> 18$	3
$d_{c - 2}^{1}$	$< 10$	1	$d_{c - 3}^{1}$	$< 5$	1	$d_{c - 4}^{1}$	$< 12$	1	$d_{c - 5}^{1}$	$< 12$	1
$d_{c - 2}^{2}$	$10 ~ 20$	2	$d_{c - 3}^{2}$	$5 ~ 15$	2	$d_{c - 4}^{2}$	$12 ~ 18$	2	$d_{c - 5}^{2}$	$12 ~ 18$	2
$d_{c - 2}^{3}$	$> 20$	3	$d_{c - 3}^{3}$	$> 15$	3	$d_{c - 4}^{3}$	$> 18$	3	$d_{c - 5}^{3}$	$> 18$	3
$d_{c - 6}^{1}$	$< 10$	1	$d_{c - 7}^{1}$	$< 5$	1	$d_{c - 8}^{1}$	$< 12$	1	${Em}_{1}^{1}$	anger (weak)	1
$d_{c - 6}^{2}$	$10 ~ 20$	2	$d_{c - 7}^{2}$	$5 ~ 15$	2	$d_{c - 8}^{2}$	$12 ~ 18$	2	${Em}_{1}^{2}$	anger (middle)	2
$d_{c - 6}^{3}$	$> 20$	3	$d_{c - 7}^{3}$	$> 15$	3	$d_{c - 8}^{3}$	$> 18$	3	${Em}_{1}^{3}$	anger (strong)	3
${Em}_{2}^{1}$	surprise (weak)	1	${Em}_{3}^{1}$	fear (weak)	1	${Em}_{4}^{1}$	anxiety (weak)	1	${Em}_{5}^{1}$	helpless (weak)	1
${Em}_{2}^{2}$	surprise (middle)	2	${Em}_{3}^{2}$	fear (middle)	2	${Em}_{4}^{2}$	anxiety (middle)	2	${Em}_{5}^{2}$	helpless (middle)	2
${Em}_{2}^{3}$	surprise (strong)	3	${Em}_{3}^{3}$	fear (strong)	3	${Em}_{4}^{3}$	anxiety (strong)	3	${Em}_{5}^{3}$	helpless (strong)	3
${Em}_{6}^{1}$	flout (weak)	1	${Em}_{7}^{1}$	relaxation (weak)	1	${Em}_{8}^{1}$	joviality (weak)	1	$Δ d^{1}$	$< 5$	1
${Em}_{6}^{2}$	flout (middle)	2	${Em}_{7}^{2}$	relaxation (middle)	2	${Em}_{8}^{2}$	joviality (middle)	2	$Δ d^{2}$	$5 ~ 15$	2
${Em}_{6}^{3}$	flout (strong)	3	${Em}_{7}^{3}$	relaxation (strong)	3	${Em}_{8}^{3}$	joviality (strong)	3	$Δ d^{3}$	$> 15$	3

Result

Parameter calibration

The 64,000 sets of valid data are selected from the above database. The population numbers is set to 400. The affinity constant $λ$ is 0.95. The adjustment factor $ε$ is 0.03.^35,36 The algorithm is run in a Matlab7.8 environment. Considering the space constraints, only the occurrence probability of driver’s different intent and iterative process are listed in anger emotion state vehicle marshaling $T_{16}$ , as shown in Figure 12.

Figure 12.

The (a) probability and (b) iterative process of driving intention appear in anger state.

Algorithm evaluation

The 32,000 sets of valid data are randomly selected from the experimental data and entered into the above algorithm. The highest driving probabilities of the algorithm are used as the final emotion stimulation result and compared with the results of the video playback and the driver’s query. The comparison results are shown in Figure 13.

Figure 13.

Comparison of algorithm results and experimental results.

Identification results of driver intentions get from alignment algorithm and experiments in anger, surprise, fear, anxiety, helplessness, flout, relaxation, and joviality emotion states as shown in Figure 13. And the probability output of the algorithm (89%, 75%, 87%, 84%, 72%, 86%, 70%, and 75%, respectively) are same as the experimental result. It can be proved that the algorithm used in this article is of higher reliability.

Discussion

The following conclusions are obtained by analyzing emotions on driving intentions in two-lane environment (Considering the space constraints, only the probability distributions of driving intentions in different emotion states in vehicle marshaling T8 are list, as shown in Figure 14.). In the same vehicle marshaling relationship, the driver will have different driving intentions in different emotional state. In different vehicle marshaling relationship, the driver will have different driving intentions in the same emotional state. Drivers in the angry state are expected have high speed. The desired spacing with surrounding car is small. They also have the higher average speed in the driving process, and the behavior decisions are more aggressive. They tend to speed up or speed up lane change. For example, the probability of speeding up lane change intention is as high as 0.333 in vehicle marshaling $T_{7}$ . Even if such behavior decisions can significantly increase the risk of driving safety, the driver still chooses to speed up lane change. When drivers are in a state of surprise, their passion for exploring the unknown things is enhanced. And they will try something new. Therefore, drivers in this emotional state tend to increase and decrease the driving speed frequently, and the lane changing frequency is higher and the driving state is unstable. For example, the probability of speed up lane change and speed down lane change intention is 0.343 and 0.301, respectively, in vehicle marshaling $T_{2}$ . When drivers are in fear emotional state, they become overcautious and the behavioral decision is more conservative because of the psychological panic. Therefore, in the emotional state, drivers’ desired speed is small, and the desired spacing with surrounding car is large, and lane changing frequency is small. That is to say, in the course of driving, drivers in the emotional state tend to choose a lower speed and keep a safe distance with the surrounding car. For example, the probability of speed down straight intention and keep the speed straight intention is 0.462 and 0.268, respectively, in vehicle marshaling $T_{5}$ . Although there are fewer vehicles around the target vehicle and the security risk is relatively small, the drivers still tends to drive at low speed. Tension or restlessness is triggered by emotional or psychological conflict, when drivers are in anxiety emotional state. They tend to frequently speed up and down, change lane in driving process, and expect to change the traffic situation as soon as possible. For example, the probability of speed up lane change and speed down lane change intention is 0.246 and 0.235, respectively, in vehicle marshaling $T_{2}$ . Drivers in the helpless emotional state tend to be lonely, not to rely on the feeling. Their speed and the desired spacing around the vehicle are larger in the vehicle driving process, and they tend to choose low speed and keep a safe distance with the vehicle around. For example, in vehicle marshaling $T_{2}$ , the drivers attempt to keep a safe distance from the forward vehicle by decreasing speed, and the probability of speed down straight intention is 0.468. One of the characteristics of drivers in flout emotional is that they feel superior to others and dislike someone or behave condescendingly. Therefore, for drivers in the state of emotional, even in a relatively dangerous environment, their driving speed and lane change frequency are also greater, and they tend to have high-speed driving. For example, in vehicle marshaling $T_{7}$ , the probability of speed up lane change intention is 0.322. For drivers in relaxation and joviality emotional states, as a result of their mental relaxation, the decisions remain in a relatively stable and reasonable state in the environment where the surrounding vehicle is more crowded. For example, the probabilities of keep the speed straight are 0.233 and 0.213 in relaxation and joviality emotional states, respectively.

Figure 14.

Probability of driving intentions in different emotional state in vehicle marshaling T_8.

Conclusions

The accurate identification of driving intention is an important guarantee for building vehicle security warning system, especially for traffic situation safety assessment subsystem. Driving emotion plays an important role in intention production and behavior decision making. Therefore, it is very important to introduce the emotion on intention guidance mechanism into the vehicle security warning system. In this article, multi-source dynamic data of human–vehicle–environment is analyzed in two-lane complex environment. Effect analysis model of emotion on driving intention based on immune algorithm is established. And the effect of emotion on driving intention is further studied. In the theory and methodology, the vacancy of the relationship study between emotion and intention in embryology is made up. In application, it provides new idea and theoretical basis to research and develop an active safety driving warning system for automobile system. But, as a preliminary study of the effect of emotion on intention, there are still many areas that need to be strengthened in this study. In the follow-up work, factors such as driver’s health condition, temperament type, blood type, occupation and other self-factors, as well as the game behavior of vehicles in interest induced area and other factors, will be further considered to reduce the experimental error. Considering the situation of driver may produce two or more emotions at the same time in the process of driving, further research on effect analysis of emotions on driving intention can focus on the more detailed driving emotion selection and definition, which would improve the accuracy of research. The theoretical basis of the auxiliary driving system of active vehicle safety is offered by the results.

Footnotes

Handling Editor: Liping Jiang

Declaration of conflicting interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This work was supported by the Joint Laboratory for Internet of Vehicles, Ministry of Education—China Mobile Communications Corporation under Project (grant number ICV-KF2018-03); the State Key Laboratory of Automotive Safety and Energy under Project (grant number KF16232); the National Natural Science Foundation of China (grant numbers 61074140, 61573009, 51508315, 51608313); Natural Science Foundation of Shandong Province (grant number ZR2014FM027); Social Science Planning Project of Shandong Province (grant number 14CGLJ27); and Project of Shandong Province Higher Educational Science and Technology Program (grant number J15LB07).

ORCID iD

Xiao-yuan Wang

References

Yao

et al . Transit network design based on travel time reliability. Transport Res C: Emer 2014; 43: 233–248.

Wang

et al . Highway management. Beijing, China: China Communications Press, 2000.

Shao

Lam

Tam

ML.

A reliability-based stochastic traffic assignment model for network with multiple user classes under uncertainty in demand. Netw Spat Econ 2006; 6: 173–204.

Shurong

Ban

Intention produce and behavior control for artificial fish based on harmonic of emotion and intelligence. Acta Autom Sinica 2007; 33: 835–839.

Ford

Merchant

Nostalgia drives donations: the power of charitable appeals based on emotions and intentions. J Advertising Res 2010; 50: 450–459.

Mohiyeddini

Pauli

Bauer

The role of emotion in bridging the intention-behaviour gap: the case of sports participation. Psychol Sport Exerc 2009; 10: 226–234.

Wang

The role of anticipated negative emotions and past behavior in individuals’ physical activity intentions and behaviors. Psychol Sport Exerc 2011; 12: 300–305.

Uskul

Kikutani

Concerns about losing face moderate the effect of visual perspective on health-related intentions and behaviors. J Exp Soc Psychol 2014; 55: 201–209.

Kleinstäuber

Gottschalk

Berking

et al . Enriching cognitive behavior therapy with emotion regulation training for patients with multiple medically unexplained symptoms (ENCERT): design and implementation of a multicenter, randomized, active-controlled trial. Contemp Clin Trials 2016; 47: 54–63.

10.

Loureiro

SMC

Roschk

. Differential effects of atmospheric cues on emotions and loyalty intention with respect to age under online/offline environment. J Retail Cons Ser 2014; 21: 211–219.

11.

Chen

Shihshuo

Tzungcheng

Nostalgic emotion, experiential value, brand image, and consumption intentions of customers of nostalgic-themed restaurants. J Bus Res 2014; 67: 354–360.

12.

Tsaur

Luoh

Syue

SS.

Positive emotions and behavioral intentions of customers in full-service restaurants: does aesthetic labor matter?

Int J Hosp Manag 2015; 51: 115–126.

13.

Wang

The impact of emotions on the intention of sustainable consumption choices: evidence from a big city in an emerging country. J Clean Prod 2016; 126: 325–336.

14.

Ladhari

Souiden

Dufour

The role of emotions in utilitarian service settings: the effects of emotional satisfaction on product perception and behavioral intentions. J Retail Cons Ser 2017; 34: 10–18.

15.

Londono

Davies

Elms

Extending the theory of planned behavior to examine the role of anticipated negative emotions on channel intention: the case of an embarrassing product. J Retail Cons Ser 2017; 36: 8–20.

16.

Harth

Leach

Kessler

Guilt, anger, and pride about in-group environmental behaviour: different emotions predict distinct intentions. J Environ Psychol 2013; 34: 18–26.

17.

Yang

Zeng

Study on the moderating effect of festival environment—based on the relationship of consumption emotion, perceived value and behavioral intention. Enterp Econ 2010; 2010: 158–160.

18.

Croog

Campus sustainability at the edges: emotions, relations, and bio-cultural connections. Geoforum 2016; 74: 108–116.

19.

Thompson

Laible

Ontai

LL.

Early understandings of emotion, morality, and self: developing a working model. Adv Child Dev Behav 2003; 31: 137–171.

20.

Cameron

Payne

Doris

JM.

Morality in high definition: emotion differentiation calibrates the influence of incidental disgust on moral judgments. J Exp Soc Psychol 2013; 49: 719–725.

21.

Chen

The effects on matching-relationship of behavior intention and outcome to children’s judgment and attribution of moral emotion. Explor Psychol 2000; 37: 44–48.

22.

Dogan

Edelbrunner

Iossifidis

. Autonomous driving: a comparison of machine learning techniques by means of the prediction of lane change behavior. In: IEEE international conference on robotics and biomimetics, Phuket, Thailand, 7–11 December 2011, pp.1837–1843. New York: IEEE.

23.

Peng

Guo

et al . Multi-parameter prediction of drivers’ lane-changing behaviour with neural network model. Appl Ergon 2015; 50: 207–217.

24.

Ohashi

Yamaguchi

Tamai

Humane automotive system using driver intention recognition. In: SICE 2004 conference, vol. 2, Sapporo, Japan, 4–6 August 2004, pp.1164–1167. New York: IEEE.

25.

Wang

Shan

et al . Prediction of bus travel time using random forests based on near neighbors. Comput-Aided Civ Inf 2018; 33: 333–350.

26.

Hao

Xiang

et al . The impact of aggressive driving behavior on driver-injury severity at highway-rail grade crossings accidents. J Adv Transport 2018; 2018: 9841498.

27.

Schmidt-Daffy

Fear and anxiety while driving: differential impact of task demands, speed and motivation. Transport Res F: Traf 2013; 16: 14–28.

28.

Jallais

Gabaude

Paire-Ficout

When emotions disturb the localization of road elements: effects of anger and sadness. Transport Res F: Traf 2014; 23: 125–132.

29.

Wang

Zhang

et al . Lane-changing model with dynamic consideration of driver’s propensity. Int J Mod Phys C 2015; 26: 1550015.

30.

Hao

Wang

et al . Developing a coordinated signal control system for urban ring road under the vehicle-infrastructure connected environment. IEEE Access 2018; 2018: 52471–52478.

31.

Wang

Liu

Zhang

Dynamic recognition model of driver’s propensity under multilane traffic environments. Discrete Dyn Nat Soc 2012; 2012: 309415.

32.

Wang

Zhang

et al . Driver’s behavior and decision-making optimization model in mixed traffic environment. Adv Mech Eng 2015; 7: 759571.

33.

Wang

Zhang

et al . Dynamic recognition of driver’s propensity based on GPS mobile sensing data and privacy protection. Math Probl Eng 2016; 2016: 1814608.

34.

Gao

Yin

Modern intelligent bionic algorithm and its application. Beijing, China: Science Press, 2011.

35.

Timmis

Ismail

Bjerknes

et al . An immune-inspired swarm aggregation algorithm for self-healing swarm robotic systems. Biosystems 2016; 146: 60–76.

36.

De Castro

. Immune, swarm, and evolutionary algorithms. Part II: philosophical comparisons. In: International conference on neural information processing, vol. 3, Singapore, 18–22 November 2002, pp.1469–1473. New York: IEEE.

37.

Tavana

Kazemi

Vafadarnikjoo

et al . An artificial immune algorithm for ergonomic product classification using anthropometric measurements. Measurement 2016; 94: 621–629.

38.

Chun

Kim

Jung

et al . Shape optimization of electromagnetic devices using immune algorithm. IEEE T Magn 1997; 33: 1876–1879.

39.

Bumgarner

Webb

Dula

CS.

Forgiveness and adverse driving outcomes within the past five years: driving anger, driving anger expression, and aggressive driving behaviors as mediators. Transport Res F: Traf 2016; 42: 317–331.

40.

Johnston

. Why We Feel The Science of Human Emotions. Quarterly Review of Biology 1999; 157: 1185–1186.

41.

Yang

Essentials of psychological experiment. Beijing, China: Beijing University Press, 1991.

42.

Nakaoka

Raksincharoensak

Nagai

. Study on forward collision warning system adapted to driver characteristics and road environment. In: International conference on control, automation and systems, Seoul, South Korea, 14–17 October 2008, pp.2890–2895. New York: IEEE.