Abstract
In recent years, the Internet of vehicles (IOV) with intelligent networked automobiles as the terminal node has gradually become the development trend of the automotive industry and the research hot spot in related fields. This is due to its characteristics of intelligence, networking, low-carbon and energy saving. Real time emotion recognition for drivers and pedestrians in the community can be utilized to prevent fatigue driving and malicious collision, keep safety verification and pedestrian safety detection. This paper mainly studies the face emotion recognition model that can be utilized for IOV. Considering the fluctuation of image acquisition perspective and image quality in the application scene of IOV, the natural scene video similar to vehicle environment and its galvanic skin response (GSR) are utilized to make the testing set of emotion recognition. Then an expression recognition model combining codec and Support Vector Machine (SVM) classifier is proposed. Finally, emotion recognition testing is completed on the basis of Algorithm 1. The matching accuracy between the emotion recognition model and GSR is 82.01%. In the process of model testing, 189 effective videos are involved and 155 are correctly identified.
Introduction
The Internet of Vehicles (IoV) promises safer roads through intelligent monitoring, but recognizing the human element—specifically emotions like fatigue, anger, or distress—is critical for preventing accidents and improving security. However, existing models often struggle in real-world vehicle environments characterized by fluctuating camera angles, poor lighting, and varying image quality. Furthermore, labeling emotions based on subjective self-reporting is often inaccurate. This research addresses these gaps by developing a robust emotion recognition model tailored for the unpredictable conditions of the IoV.
Methodology
The proposed system integrates computer vision with physiological signal processing to create a more reliable recognition framework.
Physiological Ground Truth and Preprocessing
To establish objective emotional labels, the researchers utilized Galvanic Skin Response (GSR), which measures skin conductivity changes linked to emotional arousal. To handle the inherent noise in physiological data, a Savitzky-Golay filter was applied. This digital filter fits adjacent data points to a low-order polynomial, smoothing the signal while preserving essential emotional trends.
Model Architecture
The architecture employs a convolutional codec to extract abstract features from video frames. This approach allows the model to ignore individual facial styles and focus on universal emotional markers. These features are then classified using a Support Vector Machine (SVM) with a Gaussian kernel. To ensure temporal consistency, a rule-based algorithm (Algorithm 1) aggregates frame-by-frame predictions into a final label for 15-second video segments.
Data Collection
The model was tested using "natural scene" videos of subjects under light exercise to simulate the multi-angle fluctuations and quality issues of an IoV environment. Alongside video, five channels of physiological data (including PPG, GSR, and EEG) were recorded, with GSR serving as the primary labeling reference.
Results and Discussion
The model achieved a matching accuracy of 82.01% between visual recognition and GSR-defined labels, correctly identifying the emotional state in 155 out of 189 test videos.
Key findings include:
- GSR Reliability: Using GSR peaks to categorize emotions into "Happy," "Quiet," and "Unhappy" provided a biological basis for truth, bypassing human subjectivity.
- Kernel Performance: The Gaussian kernel reached a peak training accuracy of 97.83%, outperforming polynomial and sigmoid alternatives.
- Baseline Detection: "Quiet" (calm) states were the easiest to detect, serving as a stable baseline against more high-variance expressions.
Despite its success, the model faced challenges with facial occlusion (e.g., subjects looking down) and "rich" facial styles where resting faces mimic expressions. Future work will focus on improving recognition under partial occlusion and balancing the dataset with more "Unhappy" triggers.
Conclusion
By combining convolutional feature extraction with physiological validation, this model provides a robust solution for emotion recognition in the IoV. The 82.01% accuracy in natural scenes demonstrates its potential for enhancing driver safety and community security monitoring.
Summary updated