An Intelligent System for Detecting Abnormal Behavior in Students Based on the Human Skeleton and Deep Learning.

Area of Science:

• Computer Science
• Artificial Intelligence
• Machine Learning

Background:

• Automated surveillance systems require accurate methods for detecting abnormal behavior.
• Existing methods often struggle with computational complexity and real-time processing.

Purpose of the Study:

• To develop an efficient and accurate deep learning-based method for detecting abnormal student behavior using human skeleton data.
• To reduce computational complexity and improve the accuracy of abnormal behavior identification.

Main Methods:

• Utilized the OpenPose deep learning network for extracting spatiotemporal features from human skeletons.
• Reduced feature redundancy and computational complexity using graph convolution neural networks.
• Employed a sliding window voting method to enhance classification accuracy.

Main Results:

• The proposed method achieved high accuracy, exceeding 99.50%, on both a self-built student dataset and the INRIA dataset.
• Demonstrated superior performance and practicality compared to existing abnormal behavior recognition methods.
• Achieved high processing efficiency rates, making it suitable for real-time surveillance.

Conclusions:

• The deep learning-based human skeleton analysis offers a highly accurate and efficient solution for abnormal behavior detection in surveillance.
• The integration of OpenPose, graph convolution neural networks, and sliding window voting significantly improves detection capabilities.
• This method shows great promise for enhancing security and monitoring in educational environments.