Enhancing Violent Behavior Recognition in Schools Through YOLOv8 Optimization Using LSTM with Multi-Camera Model

Violent behavior in schools poses a significant threat to the safety and well-being of students and staff. Effective recognition and prevention strategies are crucial for creating a secure educational environment. This paper presents an innovative approach to enhancing school violent behavior recognition by integrating YOLOv8 optimization with Long Short-Term Memory networks within a multi-camera surveillance model. YOLOv8, a state-of-the-art object detection framework, is optimized for real-time performance and accuracy in identifying violent actions. Including Long Short-Term Memory networks enables the system to capture temporal dependencies and improve the contextual understanding of sequential frames, thereby reducing false positives and enhancing recognition accuracy. By leveraging a multi-camera setup, the model ensures comprehensive coverage and minimizes blind spots, providing a holistic view of the monitored areas. The proposed system is evaluated on a publicly available dataset and a custom dataset collected from school environments. Experimental results demonstrate a significant improvement in the precision and recall of violent behavior recognition, underscoring the potential of this approach for real-world deployment. The findings highlight the effectiveness of combining advanced object detection algorithms with temporal analysis and multi-camera integration to create a robust and reliable violence detection system for educational institutions.