Research on the Emotional Recognition and Interactive Influence

Mechanism of the Main and Co-Pilots

Qiuyue Wang

Sydney Smart Technology College, Northeastern University, Qinhuangdao, Hebei, China

Keywords: Emotion Recognition, Emotion Linkage, Human-Vehicle Interaction, Intelligent Cockpit.

Abstract: With the evolution of intelligent cockpit technology and human-vehicle interaction systems, the ability to

recognize and regulate emotions in vehicles has increasingly become a key research direction in intelligent

driving. Existing studies mostly focus on the perception and intervention of the emotional state of the main

driver. However, in actual driving scenarios, the co-driver, as an important interactive subject, has a

significant interrelationship with the emotional state of the main driver, which may have a profound impact

on driving behavior and driving safety. In response to this relatively weak research area, this paper reviews

the latest progress in the recognition of the emotional states of the main and co-drivers, and then focuses on

the transmission mechanism and behavioral impact path of the emotional linkage between them. Finally,

based on the analysis of the challenges and gaps in current research, the research trends in the future in the

directions of linkage modeling, multimodal fusion, and human factor-adaptive interaction are discussed,

aiming to provide a theoretical basis and practical reference for building a more intelligent and collaborative

emotional perception human-vehicle interaction system.

1 INTRODUCTION

In the current era, the intelligence of smart cabins has

been significantly enhanced. The human-machine

interaction scenarios inside the vehicle are becoming

increasingly complex. Driver emotion perception and

regulation are vital for maintaining road safety and

enhancing human-vehicle interaction (Li et al., 2023;

Guo et al., 2023). Against this backdrop, emotion

recognition technology has gradually expanded from

a single visual or auditory modality to a multi-modal

fusion direction, including expressions, postures,

speech content, and some internal physiological

signals such as electroencephalogram (EEG),

electrocardiogram (ECG), electromyogram (EMG),

respiration (RSP), and temperature (T), continuously

enhancing the emotion perception capability of the

cabin system (Wu, 2023). However, current research

mainly focuses on the emotion recognition and

intervention mechanisms of the main driver. In actual

driving environments, the co-driver is also a high-

frequency interaction object, and there is a potential

interrelationship between their emotional states and

those of the main driver, which has a significant

https://orcid.org/0009-0003-3274-3087

impact on the driving behavior of the main driver and

the system response.

Unlike ordinary passengers, in real driving

scenarios, the co-driver often participates in

interactions such as navigation, reminders, and

communication, and is an important factor

influencing the emotional state of the main driver.

The emotional linkage between the main and co-

drivers may affect each other through various means

such as language, facial expressions, tone changes,

and body postures, and thereby indirectly influence

driving behavior and the intelligent response of the

vehicle system. However, current research on such

linkage phenomena is still in its infancy. Most

existing studies lack a systematic framework, and the

modeling of emotional contagion, emotional

synchronization, and their transmission paths remains

fragmented. Moreover, constructing a multi-agent

emotional linkage model in intelligent cabins still

faces many challenges, such as the alignment of

multimodal information during data collection, the

trade-off between real-time performance and

accuracy, and individual differences in emotional

Wang, Q.

Research on the Emotional Recognition and Interactive Inﬂuence Mechanism of the Main and Co-Pilots.

DOI: 10.5220/0014350100004718

Paper published under CC license (CC BY-NC-ND 4.0)

In Proceedings of the 2nd International Conference on Engineering Management, Information Technology and Intelligence (EMITI 2025), pages 239-245

ISBN: 978-989-758-792-4

239

responses, all of which limit the in-depth

advancement of related research.

Therefore, from the perspective of intelligent

cabin interaction research, this paper first reviews the

key technologies and research progress in the emotion

recognition of main and co-drivers, clarifying the

application basis in multimodal perception and

individual identification. On this basis, it further

focuses on the formation mechanism, transmission

path, and impact on driving behavior of the emotional

linkage between the main and co-drivers, revealing

the dynamic interaction characteristics of emotional

states among multiple subjects in the vehicle. In

response to the challenges and deficiencies in multi-

agent modeling, linkage perception, and response

strategy design in existing research, it deeply

analyzes the modeling methods and recognition

frameworks of emotional linkage, and proposes

design ideas for an interaction system oriented

towards multi-agent joint perception and

collaborative regulation. The related research is

conducive to improving the existing vehicle emotion

recognition system, providing theoretical support and

practical paths for enhancing the emotional

interaction capabilities of intelligent cabins and

building a safer and more stable in-vehicle emotional

ecosystem.

2 DRIVER AND CO-DRIVER

EMOTION RECOGNITION

2.1 Emotion Recognition of the Main

Driver

Within intelligent cockpit systems, accurately

identifying the driver's emotional state serves as a

critical component for enhancing both road safety and

customized user experiences. The current mainstream

research focuses on facial emotion recognition. Xiao

et al. proposed a road driver emotion recognition

method based on facial expressions called

FERDERnet. The method divides the recognition

task into three modules: first, the driver's face is

located through the face detection module; second,

the data is expanded and balanced using the

enhancement-based resampling module; in the

concluding stage, a deep convolutional neural

network, pre-trained on FER and CK+ databases then

optimized through fine-tuning, performs the driver's

emotional state classification. This method integrates

five different backbone networks and optimizes them

with an integration strategy. To verify the

effectiveness of the method, the authors constructed a

driver facial expression dataset containing a variety

of real road scenes. Experimental results show that

FERDERnet, which uses Xception as the backbone

network, outperforms the baseline network and some

advanced methods in terms of recognition accuracy

and processing efficiency, and shows excellent

performance in real road environments (Xiao et al.,

2022).

In addition, some studies have also used changes

in the driver's tone, speaking speed, volume, etc. to

judge his emotional state. Meng et al. proposed a new

deep learning architecture ADRNN (composed of

dilated convolution, residual block, BiLSTM and

attention mechanism) for speech emotion recognition

(Meng et al., 2019). This method first converts the

original speech signal into a three-dimensional Log-

Mel spectrogram as input features, and uses a dilated

convolutional network to expand the receptive field,

skip connections to retain shallow historical

information, BiLSTM to learn long-term

dependencies, and attention mechanism to further

enhance key feature extraction. In addition, the

authors introduced a combination of softmax and

center loss in the loss function to improve

classification performance. The experiment was

evaluated on two commonly used emotional speech

databases, IEMOCAP and Berlin EMODB.

Experimental findings demonstrated that the

proposed approach achieved a speaker-dependent

accuracy of 74.96% and a speaker-independent

accuracy of 69.32% in the IEMOCAP database,

which were better than the 64.74% of previous

methods. Evaluation of the EMODB dataset showed

significant improvements, yielding 90.78% accuracy

for speaker-dependent scenarios and 85.39% for

speaker-independent cases, outperforming prior

results of 88.30% and 82.82%. In addition, the

method also showed good robustness and

generalization ability in cross-corpus experiments,

achieving a recognition accuracy of 63.84% (Meng et

al., 2019). Tang et al. developed a novel end-to-end

architecture for speech emotion recognition that

combines dilated causal convolutions with context

stacking. Their design incorporates parallel

processing blocks that expand the model's receptive

field to encompass complete input sequences while

maintaining computational efficiency. Additionally,

the incorporation of context stacking enhances the

model's capacity to capture long-range dependencies.

Experiments in the regression and classification tasks

of emotion recognition show that the model achieves

better recognition performance with only about one-

third of the parameters of the current mainstream end-

EMITI 2025 - International Conference on Engineering Management, Information Technology and Intelligence

240

to-end model. In addition, the authors compared the

impact of different input representations (raw audio

vs Log-Mel spectrogram), verified the advantages of

the end-to-end learning method over hand-crafted

features, and demonstrated that the model can

effectively extract embedded features that retain

emotional information in the intermediate layers

(Tang et al., 2021).

At present, the application of driver emotion

recognition technology has gradually achieved a leap

from single modality to multimodal fusion,

combining vision, voice and even some physiological

signals to achieve a more comprehensive perception

of the driver's emotional state. Li et al. used

CogemoNet to enhance driver emotion recognition

with cognitive features. This method is different from

the traditional research method of single modality

information. The driver's facial expression is used

together with cognitive features for research. The

research team constructed a multimodal dataset

containing facial videos, cognitive feature data and

self-emotional assessment of 40 drivers. The

experimental results show that CogemoNet shows

good cross-database recognition performance on both

discrete emotion models and dimensional emotion

models, proving its effectiveness and superiority in

the driver emotion recognition task (Li et al., 2021).

Mou et al. proposed a new multimodal fusion

framework based on a convolutional long short-term

memory network (ConvLSTM) to recognize driver

emotions. This method is the first to integrate non-

invasive eye movement features, vehicle dynamics

data and environmental information with driving

context features to comprehensively model the

emotional state in driving situations. The

experimental data was collected on a highly simulated

driving simulator platform, which simulated real road

scenes through hydraulic motion systems, sound

systems and visual simulation systems. The

simulation environment supports a variety of weather

conditions (such as rain and fog), time (day and night)

and road curvature changes, inducing drivers to have

diverse emotional states. The proposed model was

verified in multiple scenarios and multiple subjects.

Following the "one scenario left out per subject"

evaluation protocol, the system accomplished mean

accuracy scores of 97.64% in valence prediction,

97.27% in arousal detection, and 96.47% in

dominance estimation; in the "leave one subject"

experiment, the accuracy rates of the three

dimensions were 88.16%, 81.65% and 85.34%

respectively, and the recall rates reached 80.97%,

72.66% and 83.62%. In addition, the ablation

experiment further revealed the different effects of

different modal features on the recognition

performance of each emotion dimension, providing a

reference for multi-task modeling (Mou et al., 2023).

2.2 Emotion Recognition of the Co-

pilot

At present, the emotion recognition research of smart

cockpit is still mainly focused on the main driver. As

the primary operator of vehicles, drivers' emotional

states have been extensively researched for their

impact on driving performance. However, as the

frequent interaction object in the car, the co-pilot is

often simplified to the identity of "passenger",

focusing on building and improving the entertainment

and comfort of the smart cockpit to meet its various

needs (Liu, Shi, & Jiang, 2021). The system assumes

that its emotions have little impact on driving, and has

not yet built an independent emotion modeling

framework for the co-pilot. As an active interactive

participant in the "third life scene", its emotional

changes may also affect the cognitive load and

psychological state of the main driver through

language communication, facial expression feedback

or even silent attitude, but related research has not

been systematically carried out.

During the driving task, the co-pilot is often in a

non-dominant task state and lacks a clear interaction

goal. Its emotional change mechanism is more hidden

and unstable. At the same time, it also lacks a

standardized annotation system and behavior label,

which makes it difficult to directly apply the existing

emotion recognition model. On the other hand, there

are more visual occlusions and perspective offsets in

the co-pilot area, which further increases the

difficulty of facial image acquisition and emotion

analysis (Yu, 2022). In voice interaction, the co-pilot

talks to the main driver more as a "cooperative

participant". There are also high individual

differences in the frequency and content structure of

their speeches, which makes it difficult for the system

to uniformly model them.

In response to the above challenges, future

research can introduce emotion recognition methods

based on body movements. Literature shows that

body movements such as head, arms, and body

postures can express emotions to a certain extent

(Tracy & Robins, 2004; Dael, Goudbeek, & Scherer,

2013), which can be applied to the emotion

recognition of co-pilots, which are more severely

restricted by light, viewing angle, etc.

Research on the Emotional Recognition and Interactive Inﬂuence Mechanism of the Main and Co-Pilots

241

3 EMOTIONAL LINKAGE AND

INFLUENCE MECHANISM

BETWEEN THE MAIN DRIVER

AND CO-DRIVER

In recent years, in the research of group intelligent

interaction, social robots and multi-user human-

computer interaction, multi-agent emotion

recognition has gradually developed into one of the

core directions of affective computing. Multi-agent

scenarios usually involve more than two interacting

subjects, whose emotional states not only change

independently, but also have complex linkage

relationships, such as emotional synchronization,

infection, and confrontation. In this closed, high-

frequency communication environment, the

emotional dynamics between the driver and the co-

pilot are more linked, and the relevant experience

provides a modeling perspective that can be used for

reference in similar scenarios.

3.1 Emotional Contagion and

Transmission Mechanism

Emotional contagion refers to the flow of emotions

between people (Van Haeringen, Gerritsen, &

Hindriks, 2023). In the closed and high-frequency

interactive space of the smart cockpit, there is often a

significant emotional resonance and emotional

contagion effect between the driver and the co-pilot.

Based on the "emotional contagion" theory of social

psychology, when one driver has obvious emotions

(such as anxiety, anger, and tension), the other driver

is easily affected unconsciously and shows a

synchronized emotional response (Pinus et al., 2025).

Emerging research indicates that the transmission of

emotions in multi-agent interactions often presents

asymmetry. In the in-car interactive scene, the main

driver's emotional state is often more likely to affect

the co-pilot due to his dominant position in vehicle

control. That is, the main driver's emotions often have

a stronger guiding effect on the co-pilot. There may

also be nonlinear dynamics, such as emotion

amplification or delayed response. In the in-car scene,

this emotional linkage may be achieved through

multimodal channels such as language, expression,

and body movements. Its transmission mechanism

has cross-modal, multi-stage, and multi-path

characteristics, which still need to be systematically

modeled.

3.2 The Mediating Effect of Emotional

Influence on Driving Behavior

The emotional linkage between the driver and the co-

pilot not only changes each other's psychological

state, but also may indirectly regulate the driving

behavior of the driver. Studies have shown that

emotions may lead to aggressive driving operations,

distracted attention or slow response, thus affecting

driving behavior (Ma, Xing, Wu, & Chen, 2024).

Driven by anger, speeding behavior will increase,

thereby increasing the possibility of traffic accidents

(Habibifar & Salmanzadeh, 2022). Fear emotions will

lead to increased heart rate, mental tension, and

reduced concentration time, resulting in a slow

response of the driver, thereby increasing the

probability of operational errors (Samuel et al., 2019).

The cognitive burden induced by stress negatively

impacts drivers' ability to maintain optimal driving

performance. (Halim & Rehan, 2020). When the co-

pilot shows anxiety or excessive intervention, the

emotional stress level of the driver will increase

significantly, which may trigger defensive or

confrontational behavior patterns. Furthermore,

emotional changes are often reflected in driving

behavior as quantifiable indicators such as steering

angle, braking frequency, and acceleration

fluctuations. Therefore, emotional linkage is not only

the object of emotion recognition, but also an

important mediating variable for understanding

driving risk status.

3.3 Linkage Recognition Framework

and Interactive System

For the emotional linkage characteristics of the driver

and the co-driver, the emotional states of multiple

people can be constructed into an emotional

propagation map through a graph neural network

(GNN) (Gao & Wang, 2024). On this basis, in the

future, we can try to build a linkage recognition

framework that integrates multimodal input. This

type of framework generally includes three key

modules: an emotion perception module that uses

signals such as voice, expression, posture, and eye

movement to extract individual emotional

characteristics; a linkage modeling module that uses

a graph neural network, a temporal neural network, or

a causal reasoning mechanism to construct the

emotional propagation path between drivers; and an

interactive feedback module that dynamically adjusts

the cabin lighting, voice assistant, or driving

assistance prompts based on the recognition results to

achieve an emotional response closed loop. For

EMITI 2025 - International Conference on Engineering Management, Information Technology and Intelligence

242

example, some intelligent cockpit systems can warn

the driver of potential tension by detecting the

emotional fluctuations of the co-driver's voice, and

reduce the volume of the audio system in the cockpit

in a timely manner to help maintain driving

concentration. The key to this type of system lies in

the dynamic understanding of the emotional

relationship between multiple subjects and the

construction of a real-time adaptive feedback

mechanism, marking an important transition from

perception to intervention in vehicle-mounted

emotional intelligence.

4 CURRENT CHALLENGES AND

RESEARCH GAPS

4.1 Difficulties in Collecting

Multi-Subject and Multi-modal

Data

In actual driving environments, the synchronous

collection of multi-subject and multi-modal data in

the car faces great technical challenges. First, due to

the physical space layout of the smart cockpit, there

may be problems such as occlusion, posture

deflection and uneven light between the driver and

the co-pilot, especially the co-pilot's side face and eye

movement features are more likely to be occluded,

affecting data integrity. Secondly, there are natural

differences in sampling frequency, timing granularity

and alignment mechanism between visual, voice,

physiological signals and other modalities, and

traditional synchronous fusion methods are difficult

to adapt to this asynchronous characteristic. In

addition, for privacy and security reasons, it is

difficult to obtain high-quality, long-term, and multi-

channel data in real environments. At present, most

studies still rely on simulation scenarios, lacking a

large-scale, natural interaction dataset of the driver

and co-pilot emotional linkage, which restricts the

model's generalizability and hinders the advancement

of research findings.

4.2 Difficulty in Dynamic Modeling of

the Emotional Linkage Mechanism

The emotional linkage between the driver and the co-

pilot has strong interpersonal interaction properties,

and its propagation process is affected by multi-factor

coupling, individual differences and emotional

asynchrony. Different from the single-person

recognition task, emotion linkage modeling needs to

deal with complex dynamic features such as emotion

source recognition, propagation direction and

intensity estimation. At present, most models focus

on individual modeling, lack relationship expression

and cross-modal and cross-time modeling capabilities,

and have not yet formed a unified linkage graph

construction method. At the same time, the

relationship between individual emotion changes and

driving behavior feedback is complex, and the causal

chain is difficult to explicitly construct, which

restricts the in-depth understanding and mechanism

mining of the linkage mechanism.

4.3 System Performance Trade-off

Between Real-Time and Accuracy

Smart cockpits place strict requirements on the real-

time performance of emotion recognition systems,

and perception and response must be completed

within sub-seconds. However, although existing deep

learning models have good expressiveness, they

consume large computing resources and are difficult

to run efficiently on vehicle-mounted edge devices,

often limited by latency and power consumption.

There is currently a lack of a unified optimization

framework for fusion model compression, inference

acceleration and modality selection mechanisms. At

the same time, multimodal data has noise and

redundancy problems, and blind fusion may reduce

recognition accuracy instead, and dynamic

scheduling mechanisms need to be developed

urgently.

5 FUTURE RESEARCH

DIRECTIONS AND TRENDS

5.1 Construction of Multimodal Graph

Model for Linkage Identification

In the future, a graph model of driver-copilot

emotional interaction can be constructed based on a

graph neural network (GNN), with the driver set as a

graph node and interaction events of different modes

modeled as edges to express "who influences whom",

"with what emotion" and "through what mode" for

transmission. Combined with time series modeling

mechanisms such as Transformer, the direction and

strength of the transmission chain can be captured,

"emotion source nodes" and "highly sensitive nodes"

can be identified, and the modeling ability of linkage

emotions in complex interaction situations can be

improved.

Research on the Emotional Recognition and Interactive Inﬂuence Mechanism of the Main and Co-Pilots

243

5.2 Design of Personalized and

Adaptive Intelligent Cockpit

Emotion Engine

For the problem of significant individual differences,

a collaborative modeling framework that integrates

individual characteristics and linkage modes should

be developed. Through methods such as federated

learning and transfer learning, personalized modeling

can be achieved under the premise of protecting

privacy; at the same time, historical linkage trajectory

analysis is introduced to achieve emotional evolution

prediction and early warning of the driver-copilot

combination. The system can actively adjust the

interaction atmosphere based on the current state,

improve positive feedback and emotional

synchronization, and enhance collaborative stability.

5.3 Development of Multi-Scenario

Highly Robust Emotion

Recognition System

Smart cockpits need to adapt to a variety of scenarios

including commuting, family travel, and long-

distance driving. The emotional interaction mode

between the driver and the co-pilot may also migrate

with the situation. Therefore, it is urgent to build a

linkage recognition system with scene adaptability.

On the one hand, a cross-scenario linkage database

can be established to train a recognition model with

universal adaptability to typical linkage modes; on the

other hand, a context-aware mechanism (such as task

intensity, in-car noise, time period, etc.) is introduced

to dynamically adjust the weights and recognition

thresholds of each modality.

5.4 Fusion of Risk Intervention

Strategies Driven by Emotion

Perception

Current intervention systems focus more on the driver

and ignore the regulatory role of the co-pilot. In the

future, a dual-subject collaborative intervention

framework based on emotional linkage relationships

can be constructed: for example, in the case of

anxiety-indifference combination, the co-pilot is

guided to intervene in communication; in the case of

double negative emotions, music, lighting and other

soothing means are used to intervene. Further

combining the linkage path with the feedback of the

intervention effect, a closed-loop system of

"recognition-response-regulation" is constructed to

improve the in-cabin emotional management and

safety assurance capabilities.

6 CONCLUSIONS

This paper focuses on the research issues of the

emotional linkage and influence mechanism between

the main and co-pilot in the intelligent cockpit

environment. With the accelerated advancement of

autonomous vehicle technologies, emotion

recognition, as a key technology to improve the in-

vehicle human-computer interaction experience and

driving safety, has received widespread attention.

However, existing research mostly focuses on the

main driver, ignoring the emotional state of the co-

pilot and its potential impact on the driving process,

and the emotional linkage relationship between the

main and co-pilot is still lacking in systematic

discussion.

To fill this gap, this paper sorts out the current

emotion recognition methods and research basis

around the emotional linkage problem between the

main and co-pilot, focusing on the emotional

propagation mechanism and behavioral mediation

effect, and proposes a corresponding recognition

framework and interactive system ideas. On this

basis, the main challenges in this field in terms of data

collection, dynamic modeling and system

performance are summarized, and it is pointed out

that current research still has problems such as

insufficient multi-agent collaborative modeling and

scene adaptability. Looking to the future, this paper

proposes key directions such as building a linkage

graph model, developing a personalized emotional

engine, improving system robustness and designing a

closed-loop intervention mechanism, which provides

theoretical support and research reference for

achieving more efficient and stable emotional

collaboration in intelligent cockpits.

REFERENCES

Dael, N., Goudbeek, M., & Scherer, K. R. (2013).

Perceived gesture dynamics in nonverbal expression of

emotion. Perception, 42(6), 642–657.

Gao, S., & Wang, Y. (2024). A review of group emotion

recognition based on images. Computers and

Modernization, (08), 98–107.

Guo, Y., Zou, Y., Xu, C., & Cao, D. (2023). A review of

emotion recognition research in smart cockpit

scenarios. In Proceedings of the 27th Annual

EMITI 2025 - International Conference on Engineering Management, Information Technology and Intelligence

244

Conference on New Network Technologies and

Applications (pp. 11–15).

Habibifar, N., & Salmanzadeh, H. (2022). Relationship

between driving styles and biological behavior of

drivers in negative emotional state. Transportation

Research Part F: Traffic Psychology and Behaviour, 85,

245–258.

Halim, Z., & Rehan, M. (2020). On identification of

driving-induced stress using electroencephalogram

signals: A framework based on wearable safety-critical

scheme and machine learning. Information Fusion, 53,

66–79.

Li, W., Cao, D., Tan, R., Shi, T., Gao, Z., Ma, J., ... & Wang,

L. (2023). Intelligent cockpit for intelligent connected

vehicles: Definition, taxonomy, technology and

evaluation. IEEE Transactions on Intelligent Vehicles,

9(2), 3140–3153.

Li, W., Zeng, G., Zhang, J., Xu, Y., Xing, Y., Zhou, R., ...

& Wang, F. Y. (2021). Cogemonet: A cognitive-

feature-augmented driver emotion recognition model

for smart cockpit. IEEE Transactions on Computational

Social Systems, 9(3), 667–678.

Liu, H., Shi, R., & Jiang, J. (2021). Discussion on the

development trend of automotive intelligent cockpit in

the 5G communication era. Journal of Guangdong

Communications Vocational and Technical College,

20(01), 33–37.

Ma, Y., Xing, Y., Wu, Y., & Chen, S. (2024). Influence of

emotions on the aggressive driving behavior of online

car-hailing drivers based on association rule mining.

Ergonomics, 67(10), 1391–1404.

Meng, H., Yan, T., Yuan, F., & Wei, H. (2019). Speech

emotion recognition from 3D log-mel spectrograms

with deep learning network. IEEE Access, 7, 125868–

125881.

Mou, L., Zhao, Y., Zhou, C., Nakisa, B., Rastgoo, M. N.,

Ma, L., ... & Gao, W. (2023). Driver emotion

recognition with a hybrid attentional multimodal fusion

framework. IEEE Transactions on Affective

Computing, 14(4), 2970–2981.

Pinus, M., Cao, Y., Halperin, E., Coman, A., Gross, J. J., &

Goldenberg, A. (2025). Emotion regulation contagion

drives reduction in negative intergroup emotions.

Nature Communications, 16(1), 1387.

Samuel, O., Walker, G., Salmon, P., Filtness, A., Stevens,

N., Mulvihill, C., ... & Stanton, N. (2019). Riding the

emotional roller-coaster: Using the circumplex model

of affect to model motorcycle riders’ emotional state-

changes at intersections. Transportation Research Part

F: Traffic Psychology and Behaviour, 66, 139–150.

Tang, D., Kuppens, P., Geurts, L., & van Waterschoot, T.

(2021). End-to-end speech emotion recognition using a

novel context-stacking dilated convolution neural

network. EURASIP Journal on Audio, Speech, and

Music Processing, 2021(1), 18.

Tracy, J. L., & Robins, R. W. (2004). Show your pride:

Evidence for a discrete emotion expression.

Psychological Science, 15(3), 194–197.

Van Haeringen, E. S., Gerritsen, C., & Hindriks, K. V.

(2023). Emotion contagion in agent-based simulations

of crowds: A systematic review. Autonomous Agents

and Multi-Agent Systems, 37(1), 6.

Wu, L. (2023). Research on driver anger emotion

recognition and regulation methods (Master’s thesis,

Chongqing University).

Xiao, H., Li, W., Zeng, G., Wu, Y., Xue, J., Zhang, J., ... &

Guo, G. (2022). On-road driver emotion recognition

using facial expression. Applied Sciences, 12(2), 807.

Yu, M. (2022). Research on the current status and

development trend of automobile intelligent cockpit

design. Engineering Management, 3(5), 187–189.

Research on the Emotional Recognition and Interactive Inﬂuence Mechanism of the Main and Co-Pilots

245