UTFacultiesEEMCSDisciplines & departmentsPSEducationMaking cycling safer: Exploring deep learning for cyclist emotion recognition

Making cycling safer: Exploring deep learning for cyclist emotion recognition

Making cycling safer: Exploring deep learning for cyclist emotion recognition

             

Problem Statement:

Even though The Netherlands is considered as the number 1 cycling country, there are signs that cycling in the country is getting more and more dangerous. The number of cyclists that suffer from severe injury in traffic accidents is increasing – in recent years with almost 30%. And an experiment in Utrecht showed that a substantial amount of woman feels unsafe when cycling alone at night-time.

Understanding cyclist emotions can help for implementing interventions to improve road safety. Wearable wristbands have been used in many studies for emotion recognition. For example, several studies showed the potential of using wristband data to understand how much stress cyclists experience. Also, libraries for physiological data analysis are available. However, there is no widespread consensus yet about how to use advanced data processing and analysis methods for getting actionable insights from multimodal cyclist emotion data.

Task:

The main objective for the student is to develop and evaluate a suitable preprocessing and deep learning pipeline for analysing multimodal sensor data of emotions during bicycle rides. The prospective student is expected to start with a brief planning and literature review. Data has been collected already, and more data will be become available soon. The challenges that are to be investigated and resolved pertain to data structuring, cleaning, data fusion, time windowing, feature engineering, model training and optimization, and result interpretation. Knowledge from signal processing, time series analysis, and frequency domain analysis is valuable. Earlier work by other students and researchers who worked on a similar topic provides starting points. The student is expected to deliver reusable code, a final presentation, and a comprehensive report.

Work:

25% theory, 50% practical, 25% report

Contact:

Mario Boot – m.r.boot@utwente.nl