Register now After registration you will be able to apply for this opportunity online.
This opportunity is not published. No applications will be accepted.
Master Thesis at ETH Zurich in Applied Computer Science: Development of an ML Algorithm for Drunk Driving Detection
Driver state detection systems will become mandatory in many countries in this decade. In this thesis, you will use a unique multi-sensor dataset with 55 drivers collected by us in a in a real car on a test track to develop a ML drunk driving detection algorithm!
Keywords: applied computer science; driver state detection; data analytics; explainable machine learning;
The improvement of vehicle technologies has equipped cars with new technologies which enable many applications. Accordingly, car driving has never been safer - yearly reports show a continuous decrease in accident reports. These improvements are attributed to new driver assistance systems and higher safety standards that were recently developed.
However, alcohol remains a major accident source and estimates see even higher numbers than the ones officially reported. Yet, there is no technology in sight to mitigate these risks. It is therefore worth investigating how driving under the influence of alcohol can be prevented.
**Important note**: We favor a flexible, agile, and pragmatic working environment. Ideally, the written thesis that the student hands in for grading will be a (high-quality) paper, that can be published after the end of the project (student will be included as author). Writing a traditional monographic thesis **is not demanded** and might only be done if the student’s department/university formally requires it. A final presentation can serve to collect bonus points for the thesis. Kickoff and intermediate presentations are not required.
The improvement of vehicle technologies has equipped cars with new technologies which enable many applications. Accordingly, car driving has never been safer - yearly reports show a continuous decrease in accident reports. These improvements are attributed to new driver assistance systems and higher safety standards that were recently developed.
However, alcohol remains a major accident source and estimates see even higher numbers than the ones officially reported. Yet, there is no technology in sight to mitigate these risks. It is therefore worth investigating how driving under the influence of alcohol can be prevented.
**Important note**: We favor a flexible, agile, and pragmatic working environment. Ideally, the written thesis that the student hands in for grading will be a (high-quality) paper, that can be published after the end of the project (student will be included as author). Writing a traditional monographic thesis **is not demanded** and might only be done if the student’s department/university formally requires it. A final presentation can serve to collect bonus points for the thesis. Kickoff and intermediate presentations are not required.
In your thesis, you will use a unique multi-sensor dataset with 55 drivers collected by us in a real car on a test track to develop a drunk driving detection algorithm (TV report of data set generation: https://youtu.be/2EkZmjB2a0M ). The participants conducted driving tasks in drunk and sober states. The goal of your thesis is to detect intoxicated driving based on the recorded data. The following sensors modalities are available:
- vehicle data (sensor measurements, control commands, status messages on the CAN bus)
- eye tracking data, gas sensor data
- vital sensor data (electrocardiogram and smartwatches)
- in- vehicle radar data.
The choice of sensor modalities for your thesis will be determined based on your specific interests and the project's overall requirements.
We are interested in understanding how we can most accurately and understandable detect that a driver is drunk. Therefore, you should use state-of-the-art machine learning approaches to build robust classifieres. You should start with focussing on conventional approaches (e.g., logistic regression, tree-based models) before using more advanced neuronal networks (e.g., RNN or CNN models). Furthermore, we expect you to explain why your model outperforms other approaches and thus you should use explainable AI approaches (e.g., statistical approaches or SHAP values) to interpret your classifier.
During the thesis, you will work closely with us and will receive dedicated supervision. We are highly interested to invest time in your thesis as the topic is one on the top of our research agenda.
In your thesis, you will use a unique multi-sensor dataset with 55 drivers collected by us in a real car on a test track to develop a drunk driving detection algorithm (TV report of data set generation: https://youtu.be/2EkZmjB2a0M ). The participants conducted driving tasks in drunk and sober states. The goal of your thesis is to detect intoxicated driving based on the recorded data. The following sensors modalities are available:
- vehicle data (sensor measurements, control commands, status messages on the CAN bus) - eye tracking data, gas sensor data - vital sensor data (electrocardiogram and smartwatches) - in- vehicle radar data.
The choice of sensor modalities for your thesis will be determined based on your specific interests and the project's overall requirements.
We are interested in understanding how we can most accurately and understandable detect that a driver is drunk. Therefore, you should use state-of-the-art machine learning approaches to build robust classifieres. You should start with focussing on conventional approaches (e.g., logistic regression, tree-based models) before using more advanced neuronal networks (e.g., RNN or CNN models). Furthermore, we expect you to explain why your model outperforms other approaches and thus you should use explainable AI approaches (e.g., statistical approaches or SHAP values) to interpret your classifier.
During the thesis, you will work closely with us and will receive dedicated supervision. We are highly interested to invest time in your thesis as the topic is one on the top of our research agenda.
Do not hesitate to apply! Please contact us with your CV and your up-to-date transcripts of records (bachelor and master). If available, an overview of completed projects can be an advantage.
Robin Deuber // rdeuber@ethz.ch // Research Assistant & PhD Candidate Chair of Information Management // WEV G220 // Weinbergstrasse 56 / 58 // 8092 Zürich
Do not hesitate to apply! Please contact us with your CV and your up-to-date transcripts of records (bachelor and master). If available, an overview of completed projects can be an advantage.
Robin Deuber // rdeuber@ethz.ch // Research Assistant & PhD Candidate Chair of Information Management // WEV G220 // Weinbergstrasse 56 / 58 // 8092 Zürich