Intelligent Analysis of Abnormal Vehicle Behavior Based on a Digital Twin

Area of Science:

• Computer Vision
• Artificial Intelligence
• Traffic Analysis

Background:

• Analyzing vehicle behavior in surveillance videos is complex due to diverse anomalies and video intricacies.
• Existing methods struggle with the variety and complexity of abnormal driving events.

Purpose of the Study:

• To propose a novel intelligent framework for vehicle behavior analysis using digital twin technology.
• To enhance the detection and analysis of abnormal vehicle activities in real-world traffic scenarios.

Main Methods:

• Vehicle detection via deep learning, followed by Kalman filtering and feature matching for tracking.
• Mapping tracked vehicles into a Unity game engine-based digital twin virtual scene.
• Behavior analysis within the digital twin using customized detection conditions and secondary scene reconstruction.

Main Results:

• The framework achieves a detection rate comparable to state-of-the-art abnormal event detection systems.
• Experimental validation using real traffic camera videos confirms the system's performance.
• Demonstrated usability, generalization, and effectiveness in analyzing vehicle behavior.

Conclusions:

• The proposed digital twin framework offers a robust solution for intelligent vehicle behavior analysis.
• The system provides a versatile platform for both real-time analysis and post-event investigation.
• This approach significantly advances the capabilities in detecting and understanding abnormal driving patterns.