Personalized trajectory inference framework integrating driving behavior recognition and temporal dependency...

Area of Science:

• Intelligent Transportation Systems
• Machine Learning
• Computer Vision

Background:

• Vehicle trajectory prediction is crucial for autonomous driving and advanced driver-assistance systems.
• Existing models often overlook the significant impact of individual driving styles on vehicle movement patterns.
• Accurate prediction requires understanding and incorporating behavioral nuances.

Purpose of the Study:

• To develop a novel model, the Driving Style-Tri Channel Trajectory Model (DS-TCTM), for enhanced vehicle trajectory prediction.
• To integrate driving style recognition into trajectory prediction for improved accuracy and safety.
• To demonstrate the model's effectiveness compared to existing baseline methods.

Main Methods:

• Data preprocessing with kinematics feature extraction.
• Driving style classification using acceleration variation, time headway, K-Means++, and K-neighbor Gaussian mixture model (K-GMM).
• Personalized trajectory prediction via a multi-level neural architecture with style-specific sub-networks.

Main Results:

• Achieved a mean Root Mean Square Error (RMSE) of 4.46 and Negative Log-Likelihood (NLL) of 3.89.
• Demonstrated a 35.8% error reduction after hyperparameter optimization.
• Showcased superior performance, especially in long-term predictions, compared to LSTM, Social-LSTM, and Convolutional-Social-LSTM models.

Conclusions:

• The DS-TCTM effectively captures the influence of driving styles on trajectory patterns.
• The model offers reliable prediction enhancements for vehicle safety systems.
• This methodology advances personalized trajectory modeling for intelligent transportation applications.