A Multifaceted Student Activity Recognition Framework from Video Surveillance Cameras
S. Anitha, R. Anita Jasmine, J. Kavitha, L. Sharmila, R. Roselinkiruba, S. Sowmyayani · 2024
The objective of this research is to identify depressed students in the school campus using CCTV footage and notify the class teachers via messages. pressure is a kingdom of strain, whether or not it be intellectual or physical. it may result from something that frustrates, incenses, or unnerves you in an occasion or thinking. In this research, we propose an algorithm for accurate emotion detection using the RAVDESS data set, with a focus on stress classification from facial expressions such as anger, calmness, happiness, disgust, and surprise. A multi modal feature fusion method is adapted for stress prediction that relies on speech and facial information. To enhance video frames, we apply Contrast Limited Adaptive Histogram Equalization (CLAHE), preserving essential structural features and improving the detection of subtle facial cues linked to stress. By employing different grid sizes (8x8 and 16x16), the method captures micro-expressions and facial tensions more effectively. Histogram equalization is also used to normalize illumination, ensuring robust performance under various lighting conditions. Experimental results show that while traditional machine learning models like SVM demonstrated reasonable accuracy (0.791), other methods such as Random Forest and Gradient Boosting struggled with high- dimensional, complex data, achieving lower quality. In contrast, deep learning models, particularly MTCNN, outperformed these approaches, with the proposed model achieving the highest accuracy of 0.856. This highlights the effectiveness of deep learning in processing video, alongside audio data, for stress detection, making it a promising tool for emotion recognition and stress analysis.