Emotion recognition for automotive drivers using simulated driving approach
In last decade, wide range of active safety system had been installed in modern vehicles. Traction control system, auto-braking system, auto wipers and auto lighting are great inventions designed to reduce road accidents. Still, statistics indicates that accident rate in Malaysia had not been com...
Saved in:
| Main Author: | |
|---|---|
| Format: | Thesis |
| Language: | English |
| Published: |
2017
|
| Subjects: | |
| Online Access: | http://psasir.upm.edu.my/id/eprint/70998/1/FK%202017%2015%20-%20IR.pdf |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | In last decade, wide range of active safety system had been installed in modern
vehicles. Traction control system, auto-braking system, auto wipers and auto
lighting are great inventions designed to reduce road accidents. Still, statistics
indicates that accident rate in Malaysia had not been compromise despite
inclusion of these features. In year 2013, approximately 777,000 registered
vehicles were involved in road traffic crashes, with damage cost of more than 9.3
billion Ringgit Malaysia. Automobile network encompasses network between
road, vehicles and drivers. Road and vehicles had made great progress,
whereas part concerning drivers had left to be the most delicate of this network.
This study encapsulates stress and anger as prime emotion encouraging road
accident. Electrodermal Activity (EDA) and Electromyography (EMG) of
corrugator supercilli had been contemplated for neutral, stress and anger
emotion recognition. Simulated driving task with preset scenario had been
developed for emotion stimulation. Experimental data were recorded from 20
healthy subjects. Acquired EDA signals were filtered, Short-Time-Fourier-
Transformed and had mean, median and variance features extracted, on the
contrary, EMG signals were rectified, filtered and had mean, standard deviation
and root mean square computed. Recorded EDA and EMG data manifested
significant difference (p < 0.05) only between neutral-stress and neutral-anger
emotion groups. Regardless, no significant difference (p > 0.05) was perceived
between stress-anger groups.
Additionally, two-class and multi-class Support Vector Machine (SVM)
classification accompanied by cross-validation method had been dispatched to
differentiate subjects’ emotion when performing simulated driving task. Dataset
from 10 subjects were used for training and another 10 were for testing purpose
only. Classification accuracy exceeding 80% had been achieved between neutral-stress and neutral-anger groups when incorporating EDA, less than 70%
accuracy was achieved for separation between stress-anger groups. EMG
features failed to perform in view of corrugator supercilli may not be compelling
measure.
This study had incorporated new techniques (Short-Time-Fourier-Transform) for
EDA analysis, apart, it is the one of the pioneer study that utilizes EDA for anger
emotion recognition, still, classification result acquired is more preferred than
past literatures. The research can still be extended by refining signal processing
techniques for better classification accuracy and conducting real-world driving
experiment for more persuasive result. |
|---|