This project implements a hybrid clustering approach combining DBSCAN (Density-Based Spatial Clustering of Applications with Noise) and Natural Breaks (Jenks) to classify traffic states. The algorithm is designed to extract meaningful traffic patterns from speed data across road segments by determining relative speed thresholds for various traffic states and analyzing the inter-segment relationships in traffic behavior.

The main functions include:

• Determining relative speed thresholds for traffic state classification using hybrid clustering.
• Computing statistical metrics (mean, standard deviation, etc.) of relative speeds in each cluster.
• Analyzing the correlation of traffic states between adjacent road segments, including:
  • Linear regression slope and intercept.
  • Pearson correlation coefficient (R).
  • p-value of statistical significance.
  • Standard error of regression coefficients.

The deviation of the FPS of adjacent vehicles will significantly reduce the CPBA of the target vehicle, and may even render it infeasible for passages to board or alight the vehicle. To mitigate this negative impact, this paper proposes a BHA* path planning method for non-standard FPS. In this method, the FPS optimization strikes a balance between the safety distance and the allowed OAVD to enhance the CPBA. Then, the BHA* algorithm rapidly generate paths for optimized FPS without FPS planning errors. The total time for FPS optimization and path planning is within 2.5 seconds, which meets the real - time requirements of practical applications. Simulation experiments based on the parameters of real vehicles and parking scenarios, as well as comparative experiments with the HA* algorithm, were conducted to verify the effectiveness and superiority of the proposed method. This research pioneers the integration of considerations regarding the CPBA into automatic parking path planning. This database contains the implementation code and supporting data for the proposed method

Codes for our paper "HUTFormer: Hierarchical U-Net Transformer for Long-Term Traffic Forecasting". Please refer to READM.md and Replication.pdf file for more details.

The dataset focuses on vehicle conflicts on the motorway, it has both vehicle kinematics data between the ego-vehicle and neighbouring vehicles, as well as traffic-based variables. The traffic variables are collected from 500m road segments and are averaged over a 5 min period before the ego-vehicle passed at every time step. If several consecutive loop detectors were not functioning, the missing cells are stored as -1. The data uses Modified Time-To-Collision as the conflict indicator. It was reconstructed a posteriori by synchronising collected driving data with corresponding traffic data that would theoretically be available in real time. This approach ensures temporal and spatial consistency between datasets, enabling realistic simulation of real-time data collection scenarios.

Replication code for 'Domain-Enhanced Dual-Branch Model for Efficient and Interpretable Accident Anticipation'.