ABSTRACT
III
ABSTRACT
Research on Driver Anger Emotion Recognition Method Based
on Multimodal Fusion
When a driver is emotionally angry, his perception, judgment, decision-making
and execution ability will be negatively affected, resulting in a decrease in driving
ability, and is prone to traffic accidents resulting in casualties and property damage.
Therefore, studying driver anger recognition methods is essential to improve traffic
safety. Although researchers at home and abroad have carried out a lot of research
work in this field, most of them only focus on the recognition of emotions in a single
modality. Since unimodal cannot fully express emotional information, the accuracy
and robustness of identifying emotions based on unimodal are low. Therefore, this
paper establishes a multimodal fusion recognition method of driver's anger based on
ECG signals and driving behavior signals, constructs a driver's anger data set and uses
a support vector machine to classify and identify anger and calm emotions.
First, emotion-inducing material is screened. Based on the Internet platform to
screen emotion-inducing video materials, analyze the results of the emotion material
calibration test and the DES scale of the discrete emotion model, and select the video
material with the highest success rate and the highest average score of emotion
induction among 35 emotion-inducing video materials in 7 groups as the
emotion-inducing video materials used to stimulate the participants' anger and calm
emotions.
Secondly, a driving anger simulation experiment was designed and the test data
was collected. The emotion induction test was used to stimulate the participants' anger
and calm emotions and carry out driving simulation tests, and the ECG signals and
driving behavior signals under anger and calm emotions were recorded by
UC-win/Road driving simulator and BIOPAC polysomatophysiography recorder,
combined with the subjective evaluation of DES scale.
Thirdly, the relationship between ECG signals and driving behavior signals on
driving anger was analyzed and feature extraction was carried out. The collected ECG
signals and driving behavior signals were sliced and processed, the linear and
nonlinear methods were used to carry out in-depth feature mining, and the extracted
candidate features were analyzed by one-way ANOVA, and 17 effective feature