CHI 2023 paper – Leveraging driver vehicle and environment interaction: Machine learning using driver monitoring cameras to detect drunk driving

This page contains our feature engineering pipeline source code for our manuscript submitted to CHI 2023:

Leveraging driver vehicle and environment interaction: Machine learning using driver monitoring cameras to detect drunk driving

Content

This repo consists of three major parts: (i) a command line tool to record the eye tracking data from a Tobii Nano Pro, (ii) a tool to calculate eye event data, and (iii) a tool to calculate features from the eye tracking data. We will describe in the following on how to get started with this code in more detail.

Prerequisites: We recommend to use Python 3.8 and to install dependencies via pip install -U -r requirements.txt

• tobii_nano_pro_recorder: A dedicated README file in the folder explains on how to use the command line tool to record Tobii Nano Pro data.
• eye_event_classification: We use the REMoDNaV algorithm to annotate the collected eye tracking data with additional events. In config/remodnav_config.json are run-specific parameters defined. In particular, we calibrated the REMoDNaV on self-annotated eye tracking data to the current parameter settings.
• eye_feature_engineering: Our custom feature engineering pipeline to create features for the prediction of drunk drivers. Several parameters can be changed in config/feature_engineering_config.json
• prediction: Here, we provide the output of our main analysis for our paper.
• examples: In this folder, we provide a simple dataset that we recorded with the Tobii Nano Pro to test our pipeline. eye_event_classification and eye_feature_engineering can be executed with this sample data.

Citation

Please cite our paper in any published work that uses any of these resources.

BiBTeX:

TBA

ACM Ref Citation:

TBA

Contact

Please contact anonymized or anonymized for questions.