Emo-aware: an intelligent and personalized model for driver emotion and health metrics assessment in advanced driver assistance systems

Driving can be enhanced if the vehicle quickly responds to the driver’s emotions. Accurate recognition of the driver’s emotional state is essential for advanced driver assistance systems (ADAS). The existing models restrict to single-modal data and fail to acquire hidden emotions which affects the prediction accuracy.

Also, they lack real time data integration with self driving vehicles. Thus, a more robust multimodal model is needed to combine driver’s vital metrics and emotional behaviors for enhanced safety and dynamic vehicle-human interactivity. The proposed model ‘Emo-Aware’ blends real-time data from multiple sensors, including face emotion detection cameras, speech emotion detection microphones, heart rate sensors, and blood pressure sensors.

The captured data are preprocessed to remove noise, retrieve relevant features and amplify to the desired state. Further, it processes this integrated data using a CNN model with hard voting method to precisely assess driver’s sentiments. The proposed framework is validated in a simulator using benchmark multimodal datasets, demonstrating its effectiveness for driver emotion assessment and laying groundwork for future real-world ADAS deployment.

The robustness of model lies in its capability to adjust driving constraints on basis of the emotion quotient of driver thereby providing a personalized driving environment. Upon model’s implementation, there is a notable rise in accuracy and fall in the loss function. The diagonal values in the confusion matrix represent class-wise emotion detection accuracies of 0.92, 0.93, 0.91, 0.95, 0.92, and 0.86, corresponding to the emotions of disgust, fear, happiness, neutrality, sadness, and surprise, respectively.

The results for discrete functional modules are also gauged. The speech-sentiment module recorded the optimal values, 0.94, 0.93, 0.91, and 0.92, respectively, for accuracy, precision, recall, and f-score. The blood pressure module reported the MSE, RMSE, MAE, and R-squared values as 6 mmHg² and 2.6 mmHg, 2.2 mmHg, and 0.96, respectively.

Furthermore, the Face module also recorded the optimum values of 0.95, 0.94, 0.93, and 0.93 for accuracy, precision, recall, and f-score, respectively. Therefore, the model can be beneficial for advanced driver emotion detection that not only emphasizes safety but also examines the relationship between drivers and their vehicles through personalized emotionally aware interactions. The authors also thank their colleagues and reviewers for their valuable suggestions, which improved the quality of this manuscript.