The codes and data for "Lane Extraction from Trajectories at Road Intersections Based on Graph Transformer Network"

Lane Extraction from Trajectories at Road Intersections Based on Graph Transformer NetworkThe data and codes that support the findings of our study "Lane Extraction from Trajectories at Road Intersections Based on Graph Transformer Network".DataThe following data are provided in the 'Data' folder:`traj.csv`: The trajectory data used for model training and testing. Each point consist of 'trajectory_id', 'timestamp', 'longitude(x)', 'latitude(y)', 'azimuth', and 'inter_id' attributes.`trueLane.csv`: The ground truth lanes for model training and testing. Each lane includes 'geometry' and 'inter_id' attributes.CodesThis repository contains the following Python codes：`data_processing.py`: Contains the implementation of data processing and feature extraction. It includes functions related to trajectory data processing, trajectory feature extraction, and the calculation of graph node and edge features.`run_process.py`: Contains the code for executing data processing and feature extraction.`GTN.py`: Contains the implementation of Graph Transformer Network for lane extraction. It includes classes and functions related to the architecture of the Graph Transformer Network and the set-based lane extraction loss calculation.`train_GTN.py`: Contains the code for Graph Transformer Network model training.`test_GTN.py`: Contains the code for model inference and lane extraction.Running the CodeData processing and feature extractionpython run_process.pyThis step processes trajectory data, extracts graph node features and edge features, and saves them as CSV files in the `processed_data` folder.Model trainingpython train_GTN.pyThis step trains the GTN model. The trained model is saved in `model/GTN_model.pth`.Model inference and lane extractionpython test_process.pyThis step performs model inference and extracts lanes for the test intersections. The lane extraction result is saved in `result/predicted_lane.csv`.RequirementsThe codes use the following dependencies with Python 3.11networkx==3.2.1pytorch==2.0.1torch-geometric==2.5.3geopandas==1.0.1