Classification Driver Emotion with Deep Learning Method for Driver Safety Detection

The main thing that car drivers care about is driving safety. At present, deaths and material losses due to car driving accidents commonly happen. According to several studies, human emotions also influence the level of risky driver behavior, for example, angry emotions show more frequent risky driving behavior more often. Based on this problem, researchers propose the classification of driver's emotions using deep learning methods. By using deep learning methods, drivers can understand their emotions and help to obtain information of driver's emotion state. The emotion detection is usually used for driving safety systems as one of the module to detect human factors in driving. In this study, KMU-FED dataset used for the emotion classification process. In addition, preprocessing step is conducted to prepare the dataset. The dataset image is resized to $48 \times 48$. Two deep learning model used for driver emotion classification. In this study, researchers applied the GhostNet and GhostNetV2 methods to find out the results in the classification of emotions in drivers. Hyperparameter tunning is conducted to get the best predictive model performance. Training and testing steps involved dataset split of 80% and 20% for training set and testing set, respectively. Model evaluation conducted to compare between GhostNet and GhostNetV2 architectures. Furthermore, the results also compared to similar previous studies. The classification results obtained using the KMU-FED dataset achieved a maximum accuracy of 99.10% on the GhostNetV2 architecture. Other methods such as SquezeeNet with Transfer Learning yield in maximum accuracy of 95.83%. This results indicates that GhostNetV2 architecture has good performance compared to the benchmark results of previous research.