Crowdsensing Road Quality using Inertial Measurement Units on Bicycles
Problem Statement
Inertial Measurement Units can measure vibration and rotational motion, and have been used on bicycles to get insights about road quality. However, most studies only developed solutions from a single probe bicycle. Several personalized factors such as tire pressure, cyclist’s weight, bicycle type, suspension, etc. affect the vibrations of the bicycle. To enable a reliable crowdsensing mechanism for measuring road quality, we need a method agnostic to these personalized factors. We thus want to develop a model that learns and negates the personalized biases to obtain generic road quality insights.
Tasks
The IMUs provided should be fixed on the bicycle’s frame (and other locations). Collect data from multiple cyclists. Identify both generic and personalized features such that all bikes can learn from each other without any bias. Your tasks are:
- Literature study (on personalized federated learning) (10%)
- Data collection (20%)
- Training generic and personalized models (40%)
- Evaluation the Writing (30%)
You will be provided with sensor-embedded bicycles.
Contact
Deepak Yeleshetty (d.yeleshetty@utwente.nl)