A driver’s adverse emotion aroused during driving increases the risk of traffic accidents by reinforcing risky driving behaviors such as aggressive driving, speeding, and traffic violation. Previous studies have developed a model to discriminate a driver’s adverse emotions using only physiological information such as ECG (electrocardiography), SC (skin conductance), EMG (electromyography), and RSP (respiration rate); however, there is insufficient research to recognize emotions based on physiological information and driving performance together. In addition, most of previous studies did not reflect differences in individual physiological characteristics in response to adverse emotions and developed a subject-independent emotion detection model rather than a personalized model.
This study aims to develop a personalized model that can detect a driver’s adverse emotional symptoms based on individual optimal ECG and driving performance measures for emotion detection. This study consists of three steps: (1) establishment of a driver emotion evaluation experimental protocol, (2) selection of optimal ECG and driving performance measures for adverse emotional symptom detection, (3) development and validation of the detection model of a driver’s adverse emotional symptoms.
First, an emotion induction method was designed to induce a driver’s adverse emotion effectively during a driving test, and an experiment that can evaluate a driver’s emotion was conducted. Sixteen healthy participants (male:8, female:8) were recruited and three different types of emotion (1. neutral, 2. anger, 3. anxiety) were evaluated during a driving test. The adverse emotion was induced by continuously presenting keywords from individual different situations that aroused anger or anxiety in the past on the driving simulator display during driving. Next, the trends of changes in ECG (IBI and LF/HF) and driving performance measures (driving speed, steering wheel rate, and centripetal acceleration) by the three types of emotion were analyzed relatively.
Second, individual optimal measures were selected for the detection of adverse emotional symptoms during driving considering the individual heart rate characteristics and driving performance. Each ECG and driving performance measure was normalized by corresponding baseline data in order to lower variability by individual differences and different units of variables. The individual’s optimal measures were selected based on the following three criteria: (1) whether the tendency of the data change by emotion is consistent with the trend of previous studies, (2) the statistical significance of the difference in the change of each measure by three different types of emotion (p < .10), and (3) repeatability (coefficient of variation < 30%).
Third, a driver-specific emotional symptom detection model was developed based on driver’s individual optimal measures and the model performance of the model was validated. Two types of detection model (1. neutral vs. anger, 2. neutral vs. anxiety) were developed with non-linear support vector machine (SVM), and the performance of the model was evaluated by k-fold cross validation (k=3). The accuracy of the SVM model was 87.5% for anger and 79.2% for anxiety.
The driver specific detection model for adverse emotional symptoms developed in this study can be used in a system that detects and warns of adverse emotional symptoms while driving to reduce the risk of traffic accidents. However, this study has a limitation that it is based on a simulated driving experiment. In an actual driving situation, ECG and driving performance measures used in this study may be influenced by factors such as surrounding vehicles, traffic conditions, and speed limits; thus, it is necessary to conduct a driver’s emotion evaluation experiment in the real driving environment in future research. In addition, this study aimed to develop a model that can discriminate between driving only situations and the situation where emotions are aroused while driving.