Confusion matrix and minimum cross-entropy metrics based motion recognition system in the classroom

Area of Science:

• Computer Vision
• Behavioral Analysis
• Machine Learning

Background:

• Classroom aggression poses risks to student safety and learning.
• Early detection of physical altercations is crucial for timely intervention.
• Existing methods may lack real-time, automated detection capabilities.

Purpose of the Study:

• To develop and validate a motion recognition system for detecting physical aggressive behavior in students.
• To enable teachers to intervene early and mitigate potential harm.
• To provide an automated solution for monitoring classroom dynamics.

Main Methods:

• Utilized camera footage to capture student movements, processing images by removing backgrounds and generating saliency maps.
• Measured key body angles (arm-to-shoulder, shoulder-to-center) and detected inter-frame velocity.
• Employed machine learning algorithms, including confusion matrix analysis and neural networks with minimum cross-entropy, for verification.

Main Results:

• The proposed motion recognition system accurately identified physical attack behaviors in collected video data.
• Simulations demonstrated the effectiveness of the developed algorithms in distinguishing aggressive actions.
• The system achieved high accuracy in detecting predefined attack criteria.

Conclusions:

• The developed motion recognition system is effective for early detection of physical aggression in classrooms.
• This technology can support educators in maintaining a safer learning environment.
• Further research can explore integration into existing school safety systems.