Proactive safety at CVIS-enabled intersections: a framework based on high-fidelity trajectory reconstruction and...

Area of Science:

• Intelligent Transportation Systems
• Computer Vision
• Robotics

Background:

• Traditional trajectory reconstruction methods struggle with complex vehicle dynamics at intersections.
• Deep learning models are computationally expensive for real-time edge deployment.
• Accurate trajectory data is crucial for proactive intersection safety.

Purpose of the Study:

• To develop an efficient, edge-computing-enhanced framework for high-fidelity vehicle trajectory reconstruction.
• To improve dynamic risk assessment and conflict detection at intersections.
• To enable real-time proactive safety interventions in Cooperative Vehicle-Infrastructure Systems (CVIS).

Main Methods:

• A two-stage framework combining physics-informed constraints, adaptive wavelet transforms, and hybrid thresholding for trajectory reconstruction.
• A Vehicle Outline-based Conflict Algorithm (VOCA) for sensitive, timely conflict detection using spatial overlap analysis.
• Implementation and validation on an NVIDIA Jetson edge device for real-world performance assessment.

Main Results:

• Reduced acceleration fluctuations by 98.66% through effective noise suppression.
• VOCA detected 77.47% more conflicts than traditional center-point methods.
• Achieved real-time performance with processing delays under 100 ms per frame per vehicle.

Conclusions:

• The proposed framework provides an efficient solution for accurate, low-latency trajectory reconstruction and conflict warnings.
• Edge-computing enhancement enables practical, real-time application of CVIS for proactive intersection safety.
• Outline-based conflict detection significantly outperforms point-based methods in complex urban environments.