CHI2025 - Moving Beyond the Simulator: Interaction-Based Drunk Driving Detection in a Real Vehicle Using Driver Monitoring Cameras and Real-Time Vehicle Data

This repository contains the source code for the preprocessing, training, and evaluation pipelines used in the following publication.

Robin Deuber, Patrick Langer, Mathias Kraus, Matthias Pfäffli, Matthias Bantle, Filipe Barata, Florian von Wangenheim, Elgar Fleisch, Wolfgang Weinmann, and Felix Wortmann. 2025. Moving Beyond the Simulator: Interaction-Based Drunk Driving Detection in a Real Vehicle Using Driver Monitoring Cameras and Real-Time Vehicle Data. In Proceedings of the 2025 CHI Conference on Human Factors in Computing Systems (CHI '25). Association for Computing Machinery, New York, NY, USA, Article 84, 1–25. https://doi.org/10.1145/3706598.3714007

Content overview

The repo consists of three majors parts, which are subdivided into the separate folders of this repository as follows:

1. 01_eye_tracking_preprocessing: Contains the preprocessing and extraction of features from eye tracking data.
2. 02_can_data_preprocessing: Contains the preprocessing and extraction of features from the CAN bus data.
3. 03_train_and_predict: Contains the code for training and evaluating the models based on the features extracted from the eyetracking and the CAN bus data.

How to use the code

The provided code can be used to build a model for drunk driving detection based on driver monitoring camera and CAN bus data. To use the code, you have to follow the following three steps below in order.

Prerequisites: We recommend to use Python >= 3.9.

1. Preprocessing eye tracking data

The first step is to preprocess the eye tracking data to extract and aggregate eye tracking features.

Adapting the config_preprocessing.yml

Inside the folder 01_eye_tracking_preprocessing, you can find the config_processing.yml. It contains some settings, such as the selected probands and the folders for the raw data and processing output. Change the following two lines you require, to load your own data:

raw_input_directory: '/test_track'
preprocessed_output_directory: '/test_track_processed/'

Adapting the config_aggregation.yml

Inside the folder 01_eye_tracking_preprocessing, you can find the config_aggregation.yml. It contains some settings, such as the selected probands and the folders for the raw data and processing output. Change the following two lines you require, to load your own data. Please note that these should be the same paths as used for the preprocessing!

data_directory: '/test_track'
data_directory_processed: '/test_track_processed/'

Installing the requirements

Please install the python requirements using the command below:

pip install -r 01_eye_tracking_preprocessing/requirements.txt

Running the code for processing and aggregating eye tracking data

First, change the directory of your terminal to the eye tracking subfolder:

cd 01_eye_tracking_preprocessing

Then, run the processing:

python 01_run_processing.py

Finally, run the aggregation to extract the features from the processed data:

python 02_run_aggregation.py

2. Preprocessing CAN data

The second step is to preprocess the CAN bus data to extract and aggregate features, such as brake and acceleration behavior and steering wheel movement.

Adapting the config_processing.yml

Just as for the eyetracking data, you have to adopt the configs for the CAN data to specify the paths to your raw data.

Inside the folder 02_can_data_preprocessing, you can find the config_processing.yml. Specify the paths for the data_directory and the data_output_directory accordingly. Make sure to use the same paths as used for the eyetracking data!

data_directory: '/test_track'
data_output_directory: '/test_track_processed'

Adapting the config_aggregation.yml

Inside the folder 02_can_data_preprocessing, you can find the config_aggregation.yml. Specify the paths for the data_directory and the data_output_directory accordingly. Make sure to use the same paths as the output path used for the eyetracking data! Also, add the prefix "/canlogger/" to the data_output_directory!

data_directory: '/test_track_processed'
relative_subject_output_directory: 'canlogger' # need to be same as the one in config_processing.yml
data_output_directory: '/test_track_processed/canlogger/'

Installing the requirements

Install the python requirements using the command below:

pip install -r 02_can_data_preprocessing/requirements.txt

Running the code for processing and aggregating CAN data

First, change the directory of your terminal to the can_data subfolder:

cd 02_can_data_preprocessing

Then, run the processing:

python 01_run_processing.py

Finally, run the aggregation to extract the features from the processed data:

python 02_run_aggregation.py

3. Training and evaluating the prediction model

Once the eye tracking and CAN data is processed and aggregated, you can use it to train the prediction model. Follow the instructions below.

Adapting the config_processing.yml

Inside the folder 03_train_and_predict, you can find the config_prediction.yml. Change the data_directory to point to the processed data (i.e., the output of the eyetracking and CAN preprocessing).

data_directory: 'test_track_processed'

Installing the requirements

Install the python requirements using the command below:

pip install -r 03_train_and_predict/requirements.txt

Running the code for training and evaluating the model

First, change the directory of your terminal to the eye tracking subfolder:

cd 03_train_and_predict

Then, run the training and evaluation:

python run_train_and_eval.py

Contributors

• Robin Deuber
• Patrick Langer
• Mathias Kraus
• Matthias Pfäffli
• Matthias Bantle
• Filipe Barata
• Florian von Wangenheim
• Elgar Fleisch
• Wolfgang Weinmann
• Felix Wortmann

Contact

Robin Deuber, rdeuber@ethz.ch

Citation

Please cite our work as follows:

Robin Deuber, Patrick Langer, Mathias Kraus, Matthias Pfäffli, Matthias Bantle, Filipe Barata, Florian von Wangenheim, Elgar Fleisch, Wolfgang Weinmann, and Felix Wortmann. 2025. Moving Beyond the Simulator: Interaction-Based Drunk Driving Detection in a Real Vehicle Using Driver Monitoring Cameras and Real-Time Vehicle Data. In Proceedings of the 2025 CHI Conference on Human Factors in Computing Systems (CHI '25). Association for Computing Machinery, New York, NY, USA, Article 84, 1–25. https://doi.org/10.1145/3706598.3714007

@inproceedings{10.1145/3706598.3714007,
author = {Deuber, Robin and Langer, Patrick and Kraus, Mathias and Pf\"{a}ffli, Matthias and Bantle, Matthias and Barata, Filipe and von Wangenheim, Florian and Fleisch, Elgar and Weinmann, Wolfgang and Wortmann, Felix},
title = {Moving Beyond the Simulator: Interaction-Based Drunk Driving Detection in a Real Vehicle Using Driver Monitoring Cameras and Real-Time Vehicle Data},
year = {2025},
isbn = {9798400713941},
publisher = {Association for Computing Machinery},
address = {New York, NY, USA},
url = {https://doi.org/10.1145/3706598.3714007},
doi = {10.1145/3706598.3714007},
booktitle = {Proceedings of the 2025 CHI Conference on Human Factors in Computing Systems},
articleno = {84},
numpages = {25},
keywords = {health, safety, driving, eye movement, vehicle interaction, driver monitoring camera},
series = {CHI '25}
}