For higher quality, click here to download the full video (MP4)
| versions: |
|
|---|---|
| sources: | https://github.com/AndresLaverdeMarin/Decoding-Driver-Type |
| keywords: | Traffic engineering, Automated Vehicles, Driver identification, Driver behavior, Deep learning |
| short name: | Decoding Driver Type: Unveiling Autonomous Vehicles using Trajectory Data |
| Copyright and License: | © Copyright (c) 2025 European Union. Licensed under the EUPL, Version 1.2 or – as soon they will be approved by the European Commission – subsequent versions of the EUPL (the "Licence");
You may not use this work except in compliance with the Licence.
You may obtain a copy of the Licence at: Unless required by applicable law or agreed to in writing, software distributed under the Licence is distributed on an "AS IS" basis, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the Licence for the specific language governing permissions and limitations under the Licence. |
| Datasets: | AstaZero (see References section) BJTU dataset (see References section) |
Table of Contents
This project contains all the code used and described in Human or Machine: A novel deep learning framework for autonomous driver identification based on vehicle trajectories.
Python-3.10+ is required. It requires numpy/scipy and pandas libraries with native backends.
Tip
On Windows, it is preferable to use the Anaconda distribution. To avoid possible incompatibilities with other projects
To run the deep learning models using GPU cuDNN 8.1 nvidia API is required.
Tip
If you use a different tensorflow-gpu version check TensorFlow 2 to install correct versions of the compilers and drivers.
Download the sources,
either with git, by giving this command to the terminal:
git clone https://github.com/AndresLaverdeMarin/ACC-Identification.git --depth=1
The files and folders of the project are listed below:
Decoding-Driver-Type
│
│ README.rst # Project documentation and usage guide
│ requirements.txt # Python dependencies and package versions
│ main.py # Main execution script for model training and evaluation
│ .gitignore # Git ignore patterns
│
├───data # Processed datasets and analysis results
│ │ AstaZero_data_processed.csv # AstaZero dataset with proposed features
│ │ AstaZero_data_processed_Li_et_al.csv # AstaZero dataset with Li et al. features
│ │ Bjtu_data_processed.csv # BJTU dataset with proposed features
│ │ Bjtu_data_processed_Li_et_al.csv # BJTU dataset with Li et al. features
│ │
│ └───ACC_data_analysis_results # Statistical analysis outputs
│ MMD_test_Scenario_A.png # Maximum Mean Discrepancy test visualization
│ MMD_test_Scenario_B.png # Maximum Mean Discrepancy test visualization
│ permutation_test_logs_scenario_A.pkl # Permutation test results for concept drift
│ permutation_test_logs_scenario_B.pkl # Permutation test results for concept drift
│
├───src # Source code for models and data processing
│ │ predict_biLSTM.py # Prediction script using biLSTM model
│ │ predict_Li_et_al_LSTM.py # Prediction script using Li et al. LSTM model
│ │ predict_logistic_regression.py # Prediction script using logistic regression
│ │ predict_SVM_linear.py # Prediction script using linear SVM
│ │
│ ├───data # Data preprocessing and feature extraction scripts
│ │ Li_et_al_processed_data.py # Data processing following Li et al. methodology
│ │ processed_data.py # Data processing for proposed feature set
│ │
│ ├───final_models # Pre-trained models and fitted scalers
│ │ biLSTM_5s.h5 # Trained bidirectional LSTM model (5-second window)
│ │ biLSTM.hdf5 # Alternative biLSTM checkpoint
│ │ Li_et_al_LSTM_5s.h5 # Trained Li et al. LSTM model (5-second window)
│ │ logistic_regression.pkl # Trained logistic regression classifier
│ │ SVM_linear.pkl # Trained linear SVM classifier
│ │ scaler_biLSTM_5s.pkl # Feature scaler for biLSTM model
│ │ scaler_Li_et_al_LSTM_5s.pkl # Feature scaler for Li et al. LSTM
│ │ scaler_logistic_regression.pkl # Feature scaler for logistic regression
│ │ scaler_SVM_linear.pkl # Feature scaler for linear SVM
│ │
│ ├───models # Model architecture definitions
│ │ biLSTM.py # Bidirectional LSTM architecture implementation
│ │ Li_et_al_LSTM.py # Li et al. LSTM architecture implementation
│ │ logistic_regression.py # Logistic regression with bagging ensemble
│ │ SVM_linear.py # Linear SVM with bagging ensemble
│ │
│ └───statistical_analysis # Statistical testing and concept drift analysis
│ statistical_test.ipynb # ANOVA and Tukey HSD tests for model comparison
│ concept_drift_analysis.ipynb # Concept drift detection using MMD
│
├───plots # Visualization and plotting utilities
│ plots.ipynb # Jupyter notebook for generating publication-ready figures
│
└───video # Animation and visualization for presentations
│ scene_1.py # Manim script for Hilbert-Huang Transform visualization
│ Video_abstract.mp4 # Video abstract of the research
│
└───data # Data files for video generation
data2plotACC.csv # Preprocessed ACC data for visualization
data2plotHV.csv # Preprocessed human vehicle data for visualization
This section demonstrates how to use the biLSTM model for driver type prediction based on the src/predict_biLSTM.py file.
To predict driver types using the pre-trained biLSTM model:
from pathlib import Path
from os.path import abspath, join, dirname
from tensorflow import keras
from data.processed_data import read
# Get the project path
prj_dir = Path(abspath(join(dirname(__file__), "..")))
# Load and preprocess data
X, y = read(
prj_dir.joinpath("data").joinpath("AstaZero_data_processed.csv"),
"./final_models/scaler_biLSTM_5s.pkl",
5,
)
# Load the pre-trained biLSTM model
model = keras.models.load_model(
prj_dir.joinpath("src").joinpath("final_models").joinpath("biLSTM_5s.h5")
)
# Evaluate the model
model.evaluate(X, y, verbose=1)
# Make predictions
predictions = model.predict(X)The read function loads trajectory data and applies the saved scaler transformation. The model uses a 5-second window (indicated by the 5 parameter) for sequence-based predictions to distinguish between human and autonomous drivers.
If you use this code or methodology in your research, please cite our paper:
Article Title: Human or Machine: A novel deep learning framework for autonomous
driver identification based on vehicle trajectories
Journal: IEEE Transactions on Intelligent Transportation Systems (TITS)
DOI: 10.1109/TITS.2025.3628491
Manuscript Number: T-ITS-24-09-3652
BibTeX entry:
@article{laverde2025human,
title = {Human or Machine: A novel deep learning framework for autonomous driver identification based on vehicle trajectories},
journal = {IEEE Transactions on Intelligent Transportation Systems},
year = {2025},
doi = {10.1109/TITS.2025.3628491},
note = {Manuscript Number: T-ITS-24-09-3652}
}
This project uses datasets from the following publications:
Citation Information:
Article Title: OpenACC. An open database of car-following experiments to study
the properties of commercial ACC systems
Authors: Makridis, Michail and Mattas, Konstantinos and Anesiadou, Aikaterini
and Ciuffo, Biagio
Journal: Transportation Research Part C: Emerging Technologies
Volume: 125
DOI: 10.1016/j.trc.2021.103047
Year: 2021
URL: https://www.sciencedirect.com/science/article/pii/S0968090X21000772
BibTeX entry:
@article{makridis_openacc_2021,
title = {{OpenACC}. {An} open database of car-following experiments to study the properties of commercial {ACC} systems},
volume = {125},
issn = {0968-090X},
url = {https://www.sciencedirect.com/science/article/pii/S0968090X21000772},
doi = {10.1016/j.trc.2021.103047},
journal = {Transportation Research Part C: Emerging Technologies},
author = {Makridis, Michail and Mattas, Konstantinos and Anesiadou, Aikaterini and Ciuffo, Biagio},
month = {April},
year = {2021},
pages = {103047}
}
Citation Information:
Article Title: On some experimental features of car-following behavior and how
to model them
Authors: Jiang, Rui and Hu, Mao-Bin and Zhang, H. M. and Gao, Zi-You and Jia, Bin
and Wu, Qing-Song
Journal: Transportation Research Part B: Methodological
Volume: 80
DOI: 10.1016/j.trb.2015.08.003
Year: 2015
URL: http://www.sciencedirect.com/science/article/pii/S0191261515001782
BibTeX entry:
@article{jiang_experimental_2015,
title = {On some experimental features of car-following behavior and how to model them},
volume = {80},
issn = {0191-2615},
url = {http://www.sciencedirect.com/science/article/pii/S0191261515001782},
doi = {10.1016/j.trb.2015.08.003},
journal = {Transportation Research Part B: Methodological},
author = {Jiang, Rui and Hu, Mao-Bin and Zhang, H. M. and Gao, Zi-You and Jia, Bin and Wu, Qing-Song},
month = {October},
year = {2015},
pages = {338--354}
}