Real-Time Trajectory Prediction Method for Intelligent Connected Vehicles in Urban Intersection Scenarios

Area of Science:

• Intelligent Transportation Systems
• Artificial Intelligence
• Robotics

Background:

• Intelligent connected vehicles (ICVs) are crucial for advancing transportation intelligence.
• Improving ICV trajectory prediction enhances traffic efficiency and safety.
• Existing methods often overlook dynamic spatial environmental factors.

Purpose of the Study:

• To propose a real-time trajectory prediction method for ICVs utilizing vehicle-to-everything (V2X) communication.
• To enhance the accuracy and reliability of ICV trajectory predictions.
• To provide a robust data support system for intelligent transportation decision-making.

Main Methods:

• Utilized a Gaussian mixture probability hypothesis density (GM-PHD) model to construct multidimensional ICV state datasets.
• Employed Long Short-Term Memory (LSTM) networks with enhanced input from GM-PHD output for consistent predictions.
• Integrated signal light factors and Q-Learning algorithms to augment LSTM with spatial features, complementing temporal data.

Main Results:

• The GM-PHD model demonstrated a 44.05% reduction in average error compared to LiDAR-based models (0.1181 m).
• The proposed enhanced LSTM model achieved a prediction error of 0.501 m.
• Reduced prediction error by 29.43% compared to social LSTM models using the Average Displacement Error (ADE) metric.

Conclusions:

• The proposed V2X-based real-time trajectory prediction method significantly improves accuracy for ICVs.
• The integration of GM-PHD and enhanced LSTM effectively incorporates dynamic spatial information.
• This research offers a valuable theoretical foundation and data support for intelligent traffic safety systems.